Janice Dowell is an associate professor of philosophy at Syracuse University and a Regular Distinguished Visiting Professor at the University of Edinburgh’s Eidyn Centre. Her work spans several subfields in philosophy, including metaphysics, philosophy of mind, philosophy of language, and ethics. In the philosophy of mind, she is best known for her work defining what it takes to be a physical property for the purposes of characterizing the thesis of physicalism. Her work in metaphysics has focused on methodological issues, particularly those concerning what it takes to justify metaphysical reductions. Her most recent work defends Kratzer’s canonical semantics for modal expressions against a series of challenges. Currently, she’s working on a book on deontic modals that supplements Kratzer’s basic semantics with an account of how it is modal propositions are determined at contexts of use (under contract with OUP). Her paper, “The Metaethical Insignificance of Moral Twin Earth”, received the 2014 Marc Sanders Prize for Metaethics.
Constructing and Justifying Semantic and Metasemantic Theories
Thanks so much to Meena for offering me this opportunity to discuss an issue I’ve been thinking about lately! These are a few early-stage thoughts on how we might best construct and justify semantic and metasemantic theories. My own immediate interest in this topic stems from my interest in assessing rival semantic theories for modal expressions in English, especially deontic ones. But my hope is that these thoughts are of some interest to those interested in semantic and metasemantic theorizing more broadly, including metaethicists interested in understanding the semantics of normative and evaluative expressions in English. Comments, questions, and suggestions very welcome.
Recently, there’s been a lot of really interesting work done by philosophers of language and linguists on understanding what sorts of meanings a semantic theory for some natural language, L, should assign the expressions of L. Is the content of a sentence at a context a set of worlds, a set ‘centered’ worlds, a set of probability spaces, a structured proposition, or what? In related debates, metaethicists wonder what sorts of meanings such a theory should assign L’s normative or evaluative expressions. How might we best approach these questions? What constraints, if any, do a plausible metasemantic theory place on good answers to them? Here are some preliminary thoughts on some of the constraints on constructing plausible semantic and metasemantic theories. (NB: Some of these thoughts are expressed in a forthcoming paper, The Metaethical Insignificance of Moral Twin Earth. The issues here, though, are not narrowly metaethical, but more broadly methodological ones for semantic and metasemantic theorizing.)
First, I make explicit a few of my background assumptions to keep us all on the same page:
A semantic theory for a language, L, assigns a meaning to each of L’s simple expressions and identifies rules of composition for complex expressions such that their meanings are the products of the meanings of the simples out of which they are composed, together with the rules they exemplify.
Among other things, a semantic theory should fit with and help explain data about what competent speakers are able to do with L, especially, communicate, coordinate, and collect information. This feature of semantic theories yields our first constraint on semantic theorizing: Since helping to provide such explanations is one of the central aims of semantic theorizing, let us call doing so the “Communication, Coordination, and Collection Constraint on Semantic Theories” or the “CCC Semantics Constraint”, for short.
In contrast to a semantic theory, a metasemantic theory for L is a theory that tells us why and how it is that the phonetic and orthographic sequences that make up L have the meanings our best semantic theory represents them as having.
How do these two kinds of theory relate to one another? Start with what we know:
1) There are noises we able to make, scribbles we are able to produce. Somehow, some of these noises and scribbles, those that correspond to words in the English language, acquired the meanings our best semantic theory, T, assigns to them.
2) Using a language with those meanings allows competent speakers to communicate, coordinate, and collect information.
Keeping this in mind, what should we think about the relationship between our theories? Should we think: Lucky are we! For we narrowly missed living in a world in which we spoke a language, but one which didn’t allow us to coordinate our activities or communicate and collect information. Or, rather, should we think: It’s no accident! Part of the explanation for why our expressions have the meanings that they do is that, in using expressions with those meanings, we are able to communicate, coordinate, and collect information. The latter is clearly so much more plausible than the former, that I propose the following,
CCC Metasemantics Constraint: A metasemantic theory should help us understand how the sequences that make up L acquired semantic significance such that speakers of L are able to do what they do with L’s expressions, centrally, communicate, coordinate, and collect information.
So far we’ve seen that that a standard characterization of semantic and of metasemantic theories each suggest constraints on candidate such theories for particular natural languages. I’ll now argue that there are important senses in which semantic and metasemantic theories are each contingent and empirical. Since contingent and empirical theories are justified by abduction (rather than, say, by derivation from a priori principles), these facts shed light on how best to construct and justify such theories.
In the sense reserved here, a feature of a world, w, is contingent just in case it varies across worlds considered as counterfactual, as ways w could have been. That there are albino tigers is a feature of the actual world, but not of all ways that world could have been, so it is a contingent feature of the actual world in this sense. A feature of a world w is empirical in the reserved sense just in case it varies across worlds considered as actual, as ways w could turn out to be. That water is H2O is an empirical feature of the actual world in this sense.
The phenomena to be explained by semantic and metasemantic theories—what the orthographic and phonetic sequences that correspond to L’s expressions mean and why they mean what they do—are contingent, empirical features of the actual world in these reserved senses; there are worlds considered as actual in which they mean a variety of different things and worlds considered as counterfactual in which they mean nothing at all. Had the actual world differed in relevant respects, the ability of speakers to use the signs they in fact do to coordinate, and what explained whether and why they have that ability, would itself have been different. The features of our world that support communication and coordination with natural languages are features that distinguish our world from others. This suffices to guarantee that those features are contingent and empirical. And that the phenomena to be explained by semantic and metasemantic theories are contingent and empirical suffices to make those theories contingent and empirical as well.
It’s perhaps easiest to see that this must be so in the case of semantic theories. First, recall the CCC semantics constraint, according to which the meaning assignments for a semantic theory for L must be capable of figuring in explanations of how it is L-competent speakers are able to communicate, coordinate, and collect information. To satisfy that constraint, a theory will need to be ‘cut down to human size’; that is, it will need to be one that reflects human cognitive and conative limitations, in particular, our limitations at the young age at which humans standardly learn their first language. Moreover, a theory that meets that constraint will need to reflect the environmental conditions under which humans learn that language.
This means that whether a semantic theory meets the CCC constraint is not a question that can be answered from the armchair (or by surveys that can be performed from many individual armchairs). This makes semantic theories more like other contingent, empirical theories–biological or psychological theories, for example–than like traditional philosophical theories about the nature of knowledge or morality. The former, though, are justified not by derivation from a priori principles, but by inference to the best explanation of the phenomena its central posits are posited to explain.
And that, I suggest, means that semantic notions, such as that of a meaning or of a semantic value, are theoretical notions, like the notion of a biological process as it figures in biological theorizing. Which semantic notions will figure in our best semantic theory for a natural language such as English will be determined by which ones figure in our best explanations of the phenomena it is the job of a semantic theory to partly explain, such as our ability to communicate, coordinate, and collect information. To make the connection between these ideas and the question with which I began clear: Deciding between hypotheses about what types of entities should serve as meanings in our best semantic theory’s meaning assignments will require taking into account the relevant human cognitive and conative limitations, as well as environmental conditions, that constrain what sort of entities could plausibly figure in explanations of human linguistic communication.
It’s perhaps less obvious that metasemantic theories are contingent and empirical in an important sense. An analogy with the debate in the philosophy of mind between physicalists of different stripes may be helpful for understanding the way in which I’m suggesting that this is so. Let a minimal physical duplicate of a world w be any world w’ that is an exact physical duplicate of w and has no additional features. A physicalist about the mental is committed to a certain necessary truth about the actual world, namely, that every world that is its minimal physical duplicate is a mental duplicate. But she is not committed to saying that the physical features that ground the mental properties instantiated at the actual world ground them at every world, whether considered as actual or counterfactual. For example, she is happy to allow that there are such worlds containing Lewisian Martians, who feel pain when fluid inflates their foot cavities. Such worlds differ from the actual world in containing physical features the actual world does not contain. On plausible empirical and metaphysical assumptions about what types of physical properties are possible, a world containing Lewisian Martians is a way the actual world could have been or even a way the actual world could turn out to be. This makes it an empirical and contingent matter which physical features of the actual world are the features that ground its mental truths. This part of the physicalist’s overall program, the task of settling the debate among physicalists about which are plausible candidate grounds for mental states and properties, is straightforwardly empirical. Moreover, as observed above, those candidate grounds will not be plausible candidate grounds in all ways the world could have been. For these reasons, the identification of candidate grounds will be largely the fruit of empirical investigation by neuroscientists into which contingent features the actual world has.
How do these observations help improve our understanding of the nature of metasemantic theories? Let S be the best such semantic theory for some natural language L. Metasemantic hypotheses for L are hypotheses about what grounds the meanings S assigns L’s expressions. Given the CCC Metasemantics Constraint, which grounds are plausible candidate grounds will depend in part on facts about human biology and psychology. Just as the entities a plausible semantic theory may assign as meanings to L’s expressions are constrained by human cognitive and conative limits, so too must the grounds of those meanings. (Slightly less abstractly: a meaning is an entity human beings are capable of communicating in part thanks to. Being a candidate grounds for such an entity must fit with this fact.)