Although philosophers have often held that causation is a purely descriptive notion, a growing body of experimental work on ordinary causal attributions using questionnaire methods indicates that it is heavily influenced by normative information. These results have been the subject of sceptical challenges. Additionally, those who find the results compelling have disagreed about how best to explain them. In this chapter, we help resolve these debates by using a new set of tools to investigate ordinary causal attributions—the methods of corpus linguistics. We apply both more qualitative corpus analysis techniques and the more purely quantitative methods of distributional semantics to four target questions: (a) Can corpus analysis provide independent support for the thesis that ordinary causal attributions are sensitive to normative information? (b) Does the evidence coming from corpus analysis support the contention that outcome valence matters for ordinary causal attributions? (c) Are ordinary causal attributions similar to responsibility attributions? (d) Are causal attributions of philosophers different from causal attributions we find in corpora of more ordinary language? We argue that the results of our analyses support a positive answer to each of these questions.

Masculinity seems to play a role in the recruitment and radicalization of lone-wolf terrorists and other violent extremists. In this paper, we examine multiple dimensions of masculinity in six corpora. We do so via linguistic analysis of the corpora associated with and produced by a range of groups and individuals. In particular, we analyze two corpora from each of: men’s rights groups, male supremacists, and manifestos of male domestic terrorists. Our results indicate that there are four distinct strands of thinking, language, and behavior in these groups and individuals: hegemonic masculinity, which manifests in domination of both women and other men, subordinated masculinity, which manifests in a resentful reactions to domination by other men, misogyny, which manifests in dominating, resentful, or outright hateful attitudes and actions towards women, and xenophobia, which manifests in fearful and vengeful reactions to invasion by and outsiders.

We investigate claims about the frequency of “know” made by philosophers. Our investigation has several overlapping aims. First, we aim to show what is required to confirm or disconfirm philosophers’ claims about the comparative frequency of different uses of philosophically interesting expressions. Second, we aim to show how using linguistic corpora as tools for investigating meaning is a productive methodology, in the sense that it yields discoveries about the use of language that philosophers would have overlooked if they remained in their “armchairs of an afternoon”, to use J.L. Austin’s phrase. Third, we discuss facts about the meaning of “know” that so far have been ignored in philosophy, with the aim of reorienting discussions of the relevance of ordinary language for philosophical theorizing. 

Scientific articles have semantic contents and structures that are usually quite specific to their disciplinary origins. After all, disciplines target specific subject matters, as do sub-disciplines or even specific articles within given journals. Here, we propose to characterize the semantic structure of scientific corpora by means of associative rules among articles topical contents as inferred with unsupervised text-mining algorithms. As a case study, we focus on one particular subfield of the humanities—the philosophy of science—for which we investigate the semantic topology of the complete corpus of one of its leading journals: Philosophy of Science. On the basis of a topic modeling of the corpus, we infer a set of 96 topic associative rules that characterize specific clusters of articles on the basis of how these articles combine topics in peculiar patterns. In a way, such topic associative rules can be interpreted as “topical recipes” for generating distinct types of philosophy of science articles. In addition, insights on the relative topical interconnectedness of article clusters can be gained by analyzing the topology of the overall rule network.

Which concepts count as thick? Agreement only exists for a few paradigmatic concepts like courageous and rude. Mostly, philosophers rely on their intuitions on how to distinguish thick concepts from thin concepts, descriptive concepts and value-associated concepts. In this talk, we discuss and present the results of three attempts to operationalize thick concepts using the methods of corpus analysis.

Facts are at the very core of intellectual activity, a precondition for understanding, and the central product of all human endeavours. There are two dominant theories of facts. On one hand, facts are taken to be abstract complexes of objects and properties, which have various logical and metaphysical features.  Facts, on this account, are specially structured building blocks of reality.  This compositional conception contrasts with a deflationary propositional account of facts, which takes facts to be true propositions.  On either theory, we may not modify the word “fact” with “true,” “loose,” “accurate,” “exact,” or “alternative” because of the word’s modal features. Philosophers, especially Hannah Arendt but also J.L. Austin, Peter Geach, and John Wisdom, have alerted us to a closely allied ordinary language epistemic sense of fact.  In this paper, I provide evidence using corpus analysis that the expression ‘… is a fact’ behaves like a modal conception of facts, while ‘facts’ behaves in the epistemic sense and may be subject to spectral moderation, which leaves it far more vulnerable to interpretation, politicisation, and—potentially—perversion.   

Many philosophically interesting concepts and distinctions seem to be acquired in early childhood. Statistical learning approaches provide a promising avenue to illuminating the acquisition of these aspects of cognition. In recent work, we have drawn on principles of statistical learning (in particular, the Size Principle) to provide an analysis of how children might come to think that knowledge is infallible (Nichols & Pinillos 2018) and how they might come to think that rules apply only to what an agent does and not to what the agent allows to happen (Nichols et al. 2016). For these kinds of statistical learning explanations to be adequate, it’s critical to determine the evidence available to the little learner (see, e.g., Foraker et al 2009). We used a corpus on child-directed speech (CHILDES) to assess the linguistic evidence available to the child in the relevant epistemic and normative domains.

Understanding how children develop adult-like representations is important in any domain within cognitive science. By investigating children's behavior at different ages as well as the potential information in their environment, we can make better inferences about the representations that children hold across development as well as the kinds of learning mechanisms that they use to arrive at these representations. In this talk, we'll present a case study from developmental linguistics that integrates corpus analyses with behavioral studies to ask how children acquire the subtle distinction in meaning between "know" and "think". While both of these propositional attitude verbs describe beliefs, they are not synonymous. "Know" is both veridical and factive, meaning that it only felicitously describes true beliefs about propositions which can be taken for granted by the speaker. In contrast, "think" is both non-veridical and non-factive in that it can describe false beliefs and also beliefs about propositions that we do not take for granted. 

I will describe some of my adventures in LDA topic modeling, and lessons learned, focusing especially on the growing realization that much of what has been done with topic models leaves much to be desired. I will discuss some of the techniques my collaborators and I have deployed to try to understand the stochastic variability that arises from the topic modeling process, and I will argue that a document-centered approach which (mostly) eschews the attempt to interpret individual topics, but instead focuses on the contexts in which documents are written and read, provides a better framework for presenting and thinking about the results of topic modeling.

In a novel extension of ‘evidentiary’ experimental philosophy, we explore how the study of word frequencies can help expose zombie intuitions. In philosophical thought experiments, as in ordinary discourse, our understanding of verbal case descriptions is enriched by automatic comprehension inferences. Such inferences have us routinely infer what else is also true of the cases described. These inferences are shaped by patterns in use frequencies: For instance, whether default inferences from a polysemous word influence further judgment and reasoning depends upon the relative frequencies of its different senses. We explore how simple corpus analyses – including human annotations and distributional semantic analysis – can be used to study such frequencies. We submit that pronounced frequency imbalances can lead to inappropriate inferences from rare uses of polysemous words. Our study shows that such inferences can generate zombie intuitions: intuitions that are ‘killed’ (defeated) by contextual information but kept cognitively alive by the psycholinguistic phenomenon of salience bias. Our paper will empirically develop and assess the hypothesis that Chalmers’s ‘zombie argument’ against materialism is built on zombie intuitions.

A computer-assisted conceptual analysis in texts aims at exploring the conceptual content of a specific highly theoretical lexical item used in philosophical texts. But because of the complexity of such task, the analysis cannot be realized by just using ready-made off-the-shelf computer programs. It requires much fine tuning or various conceptual formal, computational and computer models. Experiments on half a dozen philosophers have revealed some best practices and pitfalls of such a computer assisted endeavor.   

Philosophers tend to treat concepts of philosophical interest as if they were universal, but there are reasons to suspect that many might not be found in all cultures. In this study, we focus on epistemic concepts, providing a new perspective on the decade-long debate about the universality of epistemic concepts. To do so, we used the CLICS database of cross-linguistic colexifications (Rzymski et al. 2019) as our corpus, extending methods pioneered by Jackson et al. (2019) on emotion concepts. We examine the extent to which distinct epistemic concepts are colexified across different languages and language families, comparing them to emotion, color, and nature concepts. We also examine factors explaining variation and lack thereof, as well as striking patterns of variation and invariance. We discuss the extent to which these patterns of colexification indicate the presence or absence of universal conceptual schemes.

In this session I will present some of the research done in my lab using CHILDES to explore young children's early use of moral concepts--thin moral concepts in particular. I will then explore the value of CHILDES as a tool for experimental philosophy more generally, discussing a couple of different methodological techniques.