In their fascinating target article, Phillips and colleagues claim that the capacity to ascribe knowledge is a “basic” capacity that does not depend on the capacity to ascribe belief, and they review a large body of evidence in support of this claim: Non-linguistic studies with primates and infants that operationalize the capacity to ascribe knowledge and linguistic studies that examine linguistically how people assign knowledge. On their view, knowledge is conceived by humans, from adults to infants, by apes, and even by monkeys as a factive state, that is not just true belief, that can be obtained on the basis of all sensory modalities and by inference, and that contrasts with ignorance. Although the empirical evidence reviewed by Philips and colleagues is suggestive, it is also flawed, and the goal of this commentary is to highlight its main flaw.
The majority of studies reporting linguistic data reviewed by Phillips et al. were conducted in English with American participants – one of more than 6,500 languages currently spoken. As has been widely discussed in debates about the reliance on WEIRD (western, educated, industrialized, rich, and democratic) participants in psychology (Barrett, Reference Barrett2020; Henrich, Heine, & Norenzayan, Reference Henrich, Heine and Norenzayan2010; Simons, Shoda, & Lindsay, Reference Simons, Shoda and Lindsay2017), there are risks that come from inferring human universality from a small number of possibly unrepresentative cultures and languages. In this case, if there are differences in how knowledge is ascribed across cultures and languages, this poses a challenge for a universalist view of knowledge ascription.
The Geography of Philosophy Project (http://www.geographyofphilosophy.com) is exploring the generalizability of the findings in the linguistic studies discussed by Phillips and colleagues. This project brings together an international team of philosophers, psychologists, linguists, and anthropologists working on five continents to collect data about three important philosophical concepts: the concepts of knowledge, wisdom, and understanding. We study these concepts across a diversity of linguistic, cultural, social, economic, and religious settings.
Some preliminary results confirm the apparent universality of some patterns of knowledge ascription: As mentioned by Phillips and colleagues, across linguistic and cultural settings people tend to deny knowledge in at least some Gettier cases, thus viewing some forms of luck as being incompatible with knowledge (Machery et al., Reference Machery2017a, Reference Machery, Stich, Rose, Alai, Angelucci, Berniūnas and Zhu2017b); furthermore, stakes do not matter to the ascription of knowledge (Rose et al., Reference Rose2019).
But other results, which are directly relevant to the claims made by Phillips and colleagues, are not invariant across cultural and linguistic settings. Phillips and colleagues refer to the finding that English speakers are willing to ascribe knowledge of a given proposition while denying belief in it (Murray, Sytsma, & Livengood, Reference Murray, Sytsma and Livengood2013; Myers-Schulz & Schwitzgebel, Reference Myers-Schulz and Schwitzgebel2013; replicated with American participants in Kneer, Colaço, Alexander, & Machery, Reference Kneer, Colaço, Alexander and Macheryforthcoming), but this dissociation might not be universal: In our preliminary results, we didn't observe the effect in several countries, including Morocco and China. Phillips et al. also take knowledge to be factive: Knowledge that p is only ascribed if the ascriber takes p to be true. They gloss over disagreement among linguists about the factive uses of “know” in English. Although some take factivity to be a semantically required presupposition (e.g., Kiparsky & Kiparsky, Reference Kiparsky, Kiparsky, Bierwisch and Heidolph1970), others view it as a pragmatic phenomenon (e.g., Simons, Reference Simons2007; Vallauri & Masia, Reference Vallauri and Masia2018). If the latter is true, then it isn't the case that “know” in English expresses the representational capacity Phillips and colleagues have in mind. Be it as it may, we know very little about the factivity of the standard translations of “to know” in the thousands of languages ignored by Phillips and colleagues. Preliminary results suggest much variation in their factive behavior. Although we have observed factive uses in all the languages we have data for, several aspects of factivity vary across languages, including whether factivity is projected through negation, how the factive presupposition is canceled, and whether the terms standardly translated as “know” can be used to express a purely subjective state of confidence.
Non-experimental methods provide further evidence of variation in knowledge ascription. In Chartrand et al. (Reference Chartrand, Barr, Vindrola, Allen and Macheryin prep.), we examine the patterns of colexification of various epistemic lexemes, such as “know” and “understand” in English (see also Georgakopoulos, Grossman, Nikolaev, & Polis, Reference Georgakopoulos, Grossman, Nikolaev and Polisin press). Although “know” and “believe” are often translated by distinct lexemes, in some languages such as Cofán, a single word translates both English expressions. Speakers of these languages might still distinguish the concepts expressed in English by “know” and “believe,” but the single lexeme in Cofán that translates both “know” and “believe” may express an altogether different concept.
More generally, the image of knowledge representation and, more generally, of folk epistemology that emerges from our study is at odds with the universalist thrust of Philips and colleagues' article: We observe much variation in the use of “know” and its standard translations and more generally in the use of epistemic vocabulary. For instance, preliminary results suggest variation in whether knowledge is a norm of assertion.
Phillips and colleagues could respond that they are not interested in the meanings of “know” and its translations, but rather in a fundamental representational capacity that may differ from the meanings of these lexical items. However, they “treat knowledge as the ordinary thing meant when people talk about what others do or do not ‘know’.” Furthermore, if they are not interested in lexical meaning, why do they appeal to the use of “know” in linguistic studies to support their views? Alternatively, they could respond that non-linguistic infants’ and primates’ studies alleviate the need for cross-linguistic and cross-cultural data: If English speakers and non-linguistic creatures behave similarly, the simplest hypothesis is that all humans and some primates share a common representational capacity. However, simplicity cannot replace direct evidence of universality. Finally, they could respond that concerns about linguistic studies' generalizability leave untouched much of the reviewed evidence, which comes from non-linguistic studies, but without evidence we cannot assume that these results generalize to a diverse sample of human beings.
To conclude, we see no way around painstaking cross-cultural and cross-linguistic research when one is theorizing about fundamental human representational capacities.
In their fascinating target article, Phillips and colleagues claim that the capacity to ascribe knowledge is a “basic” capacity that does not depend on the capacity to ascribe belief, and they review a large body of evidence in support of this claim: Non-linguistic studies with primates and infants that operationalize the capacity to ascribe knowledge and linguistic studies that examine linguistically how people assign knowledge. On their view, knowledge is conceived by humans, from adults to infants, by apes, and even by monkeys as a factive state, that is not just true belief, that can be obtained on the basis of all sensory modalities and by inference, and that contrasts with ignorance. Although the empirical evidence reviewed by Philips and colleagues is suggestive, it is also flawed, and the goal of this commentary is to highlight its main flaw.
The majority of studies reporting linguistic data reviewed by Phillips et al. were conducted in English with American participants – one of more than 6,500 languages currently spoken. As has been widely discussed in debates about the reliance on WEIRD (western, educated, industrialized, rich, and democratic) participants in psychology (Barrett, Reference Barrett2020; Henrich, Heine, & Norenzayan, Reference Henrich, Heine and Norenzayan2010; Simons, Shoda, & Lindsay, Reference Simons, Shoda and Lindsay2017), there are risks that come from inferring human universality from a small number of possibly unrepresentative cultures and languages. In this case, if there are differences in how knowledge is ascribed across cultures and languages, this poses a challenge for a universalist view of knowledge ascription.
The Geography of Philosophy Project (http://www.geographyofphilosophy.com) is exploring the generalizability of the findings in the linguistic studies discussed by Phillips and colleagues. This project brings together an international team of philosophers, psychologists, linguists, and anthropologists working on five continents to collect data about three important philosophical concepts: the concepts of knowledge, wisdom, and understanding. We study these concepts across a diversity of linguistic, cultural, social, economic, and religious settings.
Some preliminary results confirm the apparent universality of some patterns of knowledge ascription: As mentioned by Phillips and colleagues, across linguistic and cultural settings people tend to deny knowledge in at least some Gettier cases, thus viewing some forms of luck as being incompatible with knowledge (Machery et al., Reference Machery2017a, Reference Machery, Stich, Rose, Alai, Angelucci, Berniūnas and Zhu2017b); furthermore, stakes do not matter to the ascription of knowledge (Rose et al., Reference Rose2019).
But other results, which are directly relevant to the claims made by Phillips and colleagues, are not invariant across cultural and linguistic settings. Phillips and colleagues refer to the finding that English speakers are willing to ascribe knowledge of a given proposition while denying belief in it (Murray, Sytsma, & Livengood, Reference Murray, Sytsma and Livengood2013; Myers-Schulz & Schwitzgebel, Reference Myers-Schulz and Schwitzgebel2013; replicated with American participants in Kneer, Colaço, Alexander, & Machery, Reference Kneer, Colaço, Alexander and Macheryforthcoming), but this dissociation might not be universal: In our preliminary results, we didn't observe the effect in several countries, including Morocco and China. Phillips et al. also take knowledge to be factive: Knowledge that p is only ascribed if the ascriber takes p to be true. They gloss over disagreement among linguists about the factive uses of “know” in English. Although some take factivity to be a semantically required presupposition (e.g., Kiparsky & Kiparsky, Reference Kiparsky, Kiparsky, Bierwisch and Heidolph1970), others view it as a pragmatic phenomenon (e.g., Simons, Reference Simons2007; Vallauri & Masia, Reference Vallauri and Masia2018). If the latter is true, then it isn't the case that “know” in English expresses the representational capacity Phillips and colleagues have in mind. Be it as it may, we know very little about the factivity of the standard translations of “to know” in the thousands of languages ignored by Phillips and colleagues. Preliminary results suggest much variation in their factive behavior. Although we have observed factive uses in all the languages we have data for, several aspects of factivity vary across languages, including whether factivity is projected through negation, how the factive presupposition is canceled, and whether the terms standardly translated as “know” can be used to express a purely subjective state of confidence.
Non-experimental methods provide further evidence of variation in knowledge ascription. In Chartrand et al. (Reference Chartrand, Barr, Vindrola, Allen and Macheryin prep.), we examine the patterns of colexification of various epistemic lexemes, such as “know” and “understand” in English (see also Georgakopoulos, Grossman, Nikolaev, & Polis, Reference Georgakopoulos, Grossman, Nikolaev and Polisin press). Although “know” and “believe” are often translated by distinct lexemes, in some languages such as Cofán, a single word translates both English expressions. Speakers of these languages might still distinguish the concepts expressed in English by “know” and “believe,” but the single lexeme in Cofán that translates both “know” and “believe” may express an altogether different concept.
More generally, the image of knowledge representation and, more generally, of folk epistemology that emerges from our study is at odds with the universalist thrust of Philips and colleagues' article: We observe much variation in the use of “know” and its standard translations and more generally in the use of epistemic vocabulary. For instance, preliminary results suggest variation in whether knowledge is a norm of assertion.
Phillips and colleagues could respond that they are not interested in the meanings of “know” and its translations, but rather in a fundamental representational capacity that may differ from the meanings of these lexical items. However, they “treat knowledge as the ordinary thing meant when people talk about what others do or do not ‘know’.” Furthermore, if they are not interested in lexical meaning, why do they appeal to the use of “know” in linguistic studies to support their views? Alternatively, they could respond that non-linguistic infants’ and primates’ studies alleviate the need for cross-linguistic and cross-cultural data: If English speakers and non-linguistic creatures behave similarly, the simplest hypothesis is that all humans and some primates share a common representational capacity. However, simplicity cannot replace direct evidence of universality. Finally, they could respond that concerns about linguistic studies' generalizability leave untouched much of the reviewed evidence, which comes from non-linguistic studies, but without evidence we cannot assume that these results generalize to a diverse sample of human beings.
To conclude, we see no way around painstaking cross-cultural and cross-linguistic research when one is theorizing about fundamental human representational capacities.
Financial support
This project/publication was made possible through the support of a grant from the John Templeton Foundation. The opinions expressed in this publication are those of the author(s) and do not necessarily reflect the views of the John Templeton Foundation.
Conflict of interest
None.