Reto Gubelmann

Transformer-based Pre-Trained Language Models currently dominate the field of Natural Language Inference (NLI). We first survey existing NLI datasets, and we systematize them according to the different kinds of logical inferences that are being distinguished. This shows two gaps in the current dataset landscape, which we propose to address with one dataset that has been developed in argumentative writing research as well as a new one building on syllogistic logic. Throughout, we also explore the promises of ChatGPT. Our results show that our new datasets do pose a challenge to existing methods and models, including ChatGPT, and that tackling this challenge via fine-tuning yields only partly satisfactory results.

In this chapter, I reconstruct Quine’s attempts to accommodate the privacy of stimulus meaning in his naturalized epistemology. These attempts span three decades and equally many monographs as well as a number of articles. Furthermore, I delineate his final proposal to solve it, which relies on natural selection to guarantee a preestablished harmony of innate perceptual similarity standards, and I discuss the extent to which this final solution still agrees with the basic doctrines that Quine defended in the 1960s (and never explicitly retracted). Finally, I develop an internal critique of two central aspects of Quine’s mature account, one focused on his use of natural selection, the other on his conception of pleasure through approval.

This chapter introduces the theme of the book, epistemological naturalism, explains central concepts and gives an overview on the book’s contents. In particular, I introduce my understanding of epistemological naturalism as consisting of two elements, namely naturalized epistemology and justificatory monism. Briefly, naturalized epistemology is the project of pursuing epistemology in a strictly scientific manner. Justificatory monism claims that there is only one fundamental way to justify any claim whatsoever, namely through empirical-scientific evidence. Furthermore, I discuss basic Quinean terminology, as Quine’s work is a central point of orientation throughout the book.

In this chapter, I show how Quine’s empiricist justificatory monism, his claim that the only fundamental way to verify any claim is through empirical evidence, is based on his verification holism, which in turn presupposes his holophrastic conception of empirical data. I then point out that the problems outlined in my discussion of Quine’s naturalized epistemology threaten to undermine his empiricist justificatory monism. Finally, I argue that Quine’s explication of empirical-scientific justification itself cannot be justified in the sense that it itself specifies. Given justificatory monism, this means that the explication is unjustified.

In this chapter, I discuss Maddy’s piecemeal realism, as she exemplifies it with regard to the debate on the ontological status of atoms that was provoked by Perrin’s seminal experiments in the early twentieth century. By considering the central experiments in some detail, by contrasting Maddy’s reading of these experiments with van Fraassen’s, and by reconstructing the arguments proposed by Perrin and Poincaré for the reality of the atom, I argue that the gap between empirical research on the one hand and scientific realism on the other cannot be closed.

This chapter serves three purposes. I first detail the basic doctrines of Quine’s naturalized epistemology, as he defended them from Word and Object until the 1990s (when he substantially modified his framework). These basic doctrines are his versions of physicalism, behaviorism, mechanism, and what I call Quine’s source- and checkpoint-empiricism. Then, I sketch the elegantly physicalistic account, given in Word and Object, of the infant’s first steps into language. Thereby, I focus on Quine’s concepts of observation sentence, stimulus meaning, and innate quality space. Finally, I detail a problem that led him to abandon this account, namely the fact that stimulus meanings are private.

In this chapter, I argue that there is a general problem for empiricist justificatory monism, the epistemological position that the only fundamental way to verify any claim is through empirical evidence. I call this problem irreducible justificatory pluralism: since each proponent of a given conception of justificatory monism is bound to this single conception, it is impossible that one of them could convince one of their competitors of their conception of justification. I argue that this dialectical deadlock can only be avoided by adopting scientific realism. I then introduce contemporary scientific realism, and I detail Quine’s rejection of it.

In this chapter, I critically assess the two so-called no-miracle arguments which Psillos proposes in support of scientific realism (I call them NMA and Meta-NMA). After pointing out that leveling the charge of vicious circularity against any of the two arguments comes at an unaffordably high price, I urge that the conclusion of Meta-NMA is false. Concerning NMA, I detail that it faces a formidable objection, the so-called pessimistic meta-induction, and I discuss Psillos’ response to this objection. Finally, I show that the conclusion of the second IBE of NMA—scientific realism with regard to a given scientific theory—falls clearly short of the standards for scientific theories that Psillos himself erects in the course of his response to the pessimistic meta-induction. I argue that this is a general problem of such holistic approaches, and I further corroborate this result by applying it on the arguments for scientific realism developed by Howard Sankey.

In this chapter, I discuss Tyler Burge’s account of the origins of objectivity. According to him, perceptual representation has a proto-predicational structure, represents the distal (as opposed to the proximal) stimulus and is necessarily governed by non-biological teleology. Furthermore, Burge argues that there is detailed empirical evidence for this conception of perceptual representation, as a mature scientific discipline, perceptual psychology, uses this conception in mathematically rigorous, predictively successful descriptions of the visual system of animal species. I discuss Burge’s proposal with an eye looking backward and an eye looking forward, as it were. First, the account adds to the critique of Quine’s own mature proposal by suggesting that Quine has been superseded by the progress of science. Second, I consider the merits of Burge’s account as an attempt at naturalized epistemology in its own rights.

This chapter discusses Penelope Maddy’s Second Philosophy of Logic and thereby again questions Quine’s mature naturalized epistemology. It is reasonable to do both at the same time, since Maddy’s proposal assumes that pre-linguistic infants refer to objects and attribute properties to them (a view that I call nativism). This view is diametrically opposed to Quine’s source-empiricism. I focus on the experimental paradigm on which Maddy relies to establish her nativism. This paradigm is called habituation and preferential looking, and it was pioneered by the developmental psychologist and outspoken nativist Elizabeth Spelke. My assessment of Maddy’s case includes specific considerations of experimental set-ups and the discussion of other empirical studies as well as more general methodological considerations.

Open-texture—e.g. vague, ambiguous, under-specified, or abstract terms—in regulatory documents lead to inconsistent interpretation, and are an obstacle to the automatic processing of regulation by computers. Identifying which parts of a legal text fall under open-texture is therefore a necessary requirement to make progress in automating the law. In this paper, we propose that large language models (LLMs) might provide an effective way to automatically detect open-texture in legal texts. We first investigate the obstacles by situating open-texture in the broader literature, and we test the hypothesis using two different LLMs—the proprietary gpt-3.5-turbo and the open-source llama-2-70b-chat—for the task of identifying open-texture in the General Data Protection Regulation. We evaluate their performance by asking 12 annotators to assess their output. We find, overall, that gpt-3.5-turbo overperforms llama-2-70b-chat on F 1 -scores (0.84 vs 0.67), and its high F 1 -score could make it a suitable alternative, or complement, to using human annotators. We also test the sensitivity of the findings against four further LLMs combined with six different prompts, and replicate a finding that there is low agreement between annotators when it comes to the identification of open-texture. We conclude the article by discussing the subjectivity of open-texture, the lessons to draw when testing for open-texture, and the consequences of using LLMs in the legal domain.

This article sets in with the question whether current or foreseeable transformer-based large language models (LLMs), such as the ones powering OpenAI’s ChatGPT, could be language users in a way comparable to humans. It answers the question negatively, presenting the following argument. Apart from niche uses, to use language means to act. But LLMs are unable to act because they lack intentions. This, in turn, is because they are the wrong kind of being: agents with intentions need to be autonomous organisms while LLMs are heteronomous mechanisms. To conclude, the article argues, based on structural aspects of transformer-based LLMs, that these LLMs have taken a first step away from mechanistic artificiality to autonomous self-constitution, which means that these models are (slowly) moving into a direction that someday might result in non-human, but equally non-artificial agents, thus subverting the time-honored Kantian distinction between organism and mechanism.

In this article, I develop a loosely Wittgensteinian conception of what it takes for a being, including an AI system, to understand language, and I suggest that current state of the art systems are closer to fulfilling these requirements than one might think. Developing and defending this claim has both empirical and conceptual aspects. The conceptual aspects concern the criteria that are reasonably applied when judging whether some being understands language; the empirical aspects concern the question whether a given being fulfills these criteria. On the conceptual side, the article builds on Glock’s concept of intelligence, Taylor’s conception of intrinsic rightness as well as Wittgenstein’s rule-following considerations. On the empirical side, it is argued that current transformer-based NNLP models, such as BERT and GPT-3 come close to fulfilling these criteria.

At the intersection of epistemology, metaphilosophy, and philosophy of science, this exciting new book examines the epistemic limits of empirical science. It makes a unique contribution to research on epistemological naturalism in Quine’s tradition by criticizing the position based on first-order data from empirical psychology and the history of natural science. This way, it meets the naturalist on their own ground not only regarding subject matter, but also regarding their epistemic methods. The book explores the works of a variety of philosophers in the field, including W. V. Quine, Penelope Maddy, Tyler Burge, Stathis Psillos and Howard Sankey. By carefully considering experimental results from behaviourism as well as developmental and perceptual psychology, Gubelmann finds that none of these disciplines can furnish the epistemic means to successfully naturalize the central cognitive preconditions of scientific theorizing. Furthermore, Gubelmann presents novel arguments for the claims that epistemological naturalists are committed to scientific realism, and that they are unable to defend this position. Based on these results, Gubelmann concludes that epistemology is not part of empirical science, which directly contradicts epistemological naturalism.

This article critically assesses the empirical research that leads Quine, in his posthumously published work, to abandon his empiricist principle that humans do not have any innate concepts, or knowledge. It is the same empirical research that Penelope Maddy capitalizes on to develop her own contributions to naturalized epistemology, and it has been pioneered by developmental psychologist Elisabeth Spelke. Spelke employs the method of habituation and preferential looking to argue that human infants have innate concepts, and that they have some knowledge about what can and cannot happen to physical objects. Taking into account empirical studies as well as methodological considerations, this article examines whether this research can support these strong philosophical conclusions drawn from it, finding that it likely cannot provide such support.

W. V. O. Quine is generally seen as one of the foremost empiricists of the twentieth century. For large parts of his career, the label “empiricist” is accurate; in his mature work, however, he integrated decidedly antiempiricist elements in his epistemology. From The Roots of Reference onward, he enlists natural selection and innate cognitive structures to ensure that scientific concepts have a “degree of objective validity.” From From Stimulus to Science onward, he also explains the very possibility of communication via a preestablished harmony of innate cognitive structures that is guaranteed by natural selection. This article reconstrues the reasons that compelled Quine to these commitments, and it details the development of Quine’s thinking on these topics across more than 3 decades; in particular, the article argues that recognizing that so-called stimulus meanings are private decisively shaped Quine’s views. By means of a critical evaluation, the article argues that natural selection can make plausible that scientific concepts have a degree of objective validity—if this Quinean claim is properly understood; in contrast, the article suggests, with recourse to research by Robert C. Richardson, that it is doubtful whether natural selection can underpin the preestablished harmony that Quine requires to explain communication.