This paper develops an eliminativist critique of dual character concept theories, which posit that some concepts exhibit a "double dissociation" between descriptive and normative categorization. By examining different categorization judgments, I demonstrate that double dissociation is partly better explained by context-dependent applications of broadened category standards, rather than the hybrid lexical structures posited by extant theories. Eliminativism offers a more parsimonious treatment of double dissociation, one that cautions against overreliance on ambiguous linguistic tests and emphasizes the importance of distinguishing between literal and loose categorizations. This position challenges dual character concept theories and provides methodological inputs for future experimental work in categorization.

Theorists note a distinction within the category of generics between descriptive generics that merely describe generalizations, and normative generics, that have a distinctly normative flavor. Previous proposals have analyzed the latter as speech acts that give instructions, provide advice, or prescribe appropriate behavior. We concur that these are important and central uses of normative generics, but argue that the analysis should broaden its focus to some of the wider ways that normative generics figure in discourse. Uses that give voice to derogations or to threats, are also quite central, yet the accounts alluded to above wouldn’t adequately explain why. After making the case for these uses of normative generics, our talk focuses on bringing together resources that allow us to explain them. We argue that the metalinguistic account of definitional generics best allows us to see normative generic discourse as a means of negotiating and policing normative expectations about the relevant kinds, and in turn, to analyze the relevant uses described above as derogations and/or threats.

Are the outputs generated by LLM-based chatbots---e.g. ChatGPT, Gemini, and DeepSeek---meaningful in the same way in which human-generated speech is meaningful? In response to this question, I explore and partially defend a position which I call Meaning Minimalism about LLMs. In the course of this, I place special emphasis on the question in how far LLM-based AI-systems can perform speech acts.

In recent years, several empirical studies have been conducted to investigate whether people think it is possible to lie with deceptive implicatures. While the findings are not clear cut, there is quite strong evidence that at least some cases of deceptive implicatures are considered to be lies. Moreover, it has been found that some literally true utterances that convey a false implicature are judged to be false.

In the present paper, we investigate whether the aforementioned phenomena can be generalized to different languages and cultures. Moreover, we test the degree to which human judgments are aligned with judgments elicit by artificial intelligences (in the form of LLMs such as ChatGPT). To these aims, we conducted a cross-cultural study in which participants were presented with ten cases of deceptive implicatures and asked to judge whether the speaker lied and (in a different condition) whether what she said was true or false. 3660 participants (183 per condition; 366 per country) were recruited in ten countries (USA, UK, South Africa, Spain, Chile, Mexico, Israel, Germany, China, and Japan). Surprisingly, highly similar result patterns were found between countries, with correlations between different countries being higher than correlations between humans and AI.