It's Just Semantics
Matthew Simons • Colorado College
Say you are giving the definition of ‘dog’ to a child and explain it to be a creature with four legs and fur that lives with humans. The child then points to a cat and says, “That is a dog!” You would then correct the child and tell her that what she is pointing to is a cat, not a dog. But within the parameters of the definition given, the child is correct in assuming that the animal pointed to is a dog. After all, cats also have four legs and fur and live with humans. There is no way for her to tell the difference based on your definition. This is an example of “overextension” in early childhood language development: the use of a word to refer to things that do not fit within its definition. But how do children fix their extension errors? How do we begin to use words correctly and understand exactly what they mean?
One would quite naturally assume that the words we use in our languages have meanings. And for a word to have meaning, there must be correct and incorrect ways of using that word. What determines how we use a word correctly? One possible answer is, “The meaning of a linguistic expression determines what counts as correctly or incorrectly applying the expression in particular instances.” This concept is called semantic normativity, and at first glance, it almost seems like a truism. The semantic normativity thesis should seem fairly natural to most language users; after all, when most people are in a dispute over what a word means, they look to the dictionary to define that word for them. A definition gives us the meaning of a word which then determines how to use that word correctly. Definitions can be viewed as rules for how to correctly use a word. And this idea that correctness is determined by the meaning or definition of a word is semantic normativity.
The Regress and Gerrymandering Arguments
Ludwig Wittgenstein, in his book Philosophical Investigations, challenges the concept of semantic normativity. The regress and gerrymandering arguments, as they are called by Wittgenstein scholars, draw the reader’s attention to the problematic nature of semantic normativity through two different thought experiments. Let us look back to the child misinterpreting the word ‘dog’ to understand the regress argument. After the child has incorrectly identified the cat with the word ‘dog,’ you would naturally correct her and then attempt to provide a more accurate definition of ‘dog.’ You could say that a ‘dog’ is usually bigger than a cat and barks instead of meows. The issue with this definition of ‘dog’ is that the child could also misinterpret it as she did the first one. She could see a coyote and refer to it as ‘dog’ because coyotes also bark and are larger than cats. A rule that clarifies a misunderstood rule can also be misunderstood. Even a rule explaining a rule is the same kind of thing: “The regress arises in each case because each thing we introduce to determine how it should be correctly applied is merely another object of this kind.” This generates an infinite regress in which there is no rule or definition that can remove all doubt as to how to correctly use a word.
A rule is in need of some interpretation by the person attempting to understand it. And if the first rule can be misinterpreted, then the next rule implemented to clarify the first misinterpretation, can also be misinterpreted. A definition uses words to explain what another word means, but if the words in the definition are not already understood by the person learning the definition of the word, then the definition serves no purpose.
Despite the theoretical strength of the regress argument, we still seem to understand how to use words correctly. Most adult language users are not regularly confusing dogs with cats, bears, or anything else that does not fit the definition of ‘dog.’ Is this because they have fully understood the rule for how to use the word ‘dog’? Does understanding the definition of ‘dog’ allow adult language users to know all possible correct situations in which to use the word? Wittgenstein does not think so, and he uses the gerrymandering thought experiment to demonstrate that knowing the meaning of a word cannot guarantee that anybody else will understand the meaning as you do.
Say you are given the series 4, 8, 12, 16, 20 . . . How would you continue that series? What is the rule for that series? Most people would say the rule for the series is +4 to each consecutive number. But I could also continue the series like this: 25, 30, 35, 40, 45, 50, 56, 62, 68 . . . You might be inclined to call my continuation of the series incorrect. In response, I explain that I thought the series was +4 for the first four numbers, +5 for the next five numbers, +6 for the next 6 numbers, etc. It is unlikely that most people seeing the original series would extrapolate this pattern, but my rule for the series is just as correct as the +4 rule. The reason we might reject my rule for the series is that we usually believe, for a series of this kind, that “‘the right step is the one that is in accordance with the order—as it was meant.’” The creator of the series probably meant for it to be continued with the +4 rule, but the series itself cannot guarantee a shared interpretation of how it is supposed to continue correctly.
This reasoning can be applied to words as well. Take the word ‘pillow’: All our knowledge of that word is based on past experiences we have had. Somebody pointed to an object and called it a ‘pillow,’ or we saw pillows on TV or in a store. These personal experiences we have with the word ‘pillow’ give us its meaning. Each use of ‘pillow’ in our life is like a number in the series, and the meaning of ‘pillow’ is the rule for how to continue the series. There is no guarantee that I will continue to use the word ‘pillow’ in a manner that seems intuitive to you. My understanding of the definition of ‘pillow’ might be different because I learned the word through different examples than you did. When I use ‘pillow’ to refer to a chair, you might try and clarify what ‘pillow’ actually means. But when you attempt to further explain what ‘pillow’ means, you fall back into the infinite regress in which I could misinterpret all the words you use in your definition of ‘pillow.’
Family Resemblance Model
The arguments against semantic normativity present us with a difficult dilemma. It seems that a word’s meaning cannot create the rule for its correct usage, yet we usually understand when a word is being used correctly or incorrectly. So, there must be an alternative way to think about how we know the correct and incorrect usage of a word. Though Wittgenstein does not present a positive theory of language in Philosophical Investigations, he does introduce the concept of family resemblance to suggest how we might know how to use words correctly. Family resemblance is the idea that words are connected through their various uses by a string of ‘affinities and similarities’ between each different meaning or way of interpreting the word. Wittgenstein explains the concept using the word ‘game’: “Look, for example, at board-games, with their various affinities. Now pass to card-games; here you find many correspondences with the first group, but many common features drop out, and others appear. When we pass next to ball-games, much that is common is retained, but much is lost. — Are they all ‘entertaining’? Compare chess with noughts and crosses [tic tac toe].” He then continues to describe various other games: some are competitive, some involve balls, some have teams, etc. But there is nothing that is common to all games, only a web of similarities that connect all of them. Wittgenstein seems to believe that all words function like this. Fake plastic lemons, drawings of lemons, non-yellow lemons, and normal lemons all have overlapping aspects, but there are not any characteristics that every single one of them shares.
The family resemblance concept escapes the pitfalls of semantic normativity because one cannot create a rule for the correct usage of a word based on meanings which are strung together with various affinities but no central idea. For a rule or definition to govern the correct usage of a word, there needs to be at least one characteristic that is central to every possible usage of that word. In the family resemblance model, there is no single characteristic that all uses of the word ‘lemon’ share. This makes it difficult to delineate exactly which uses of the word ‘lemon’ are correct and which are incorrect because the category of ‘lemon’ does not have clear boundaries as to which things do or do not belong within the category based on some essential characteristic(s).
The family resemblance concept resolves a lot of the issues with semantic normativity and seems to explain a lot of the nuances in our language, but does the concept extend beyond the theoretical realm? Do we think of a word as a string of connections between its various iterations? The family resemblance model does not seem to comport with how I think most adult language users think about the meaning of the words they know. I do not think the word ‘lemon’ is a loose string of connections between the various ways I use the word; rather, there seems to be something that makes a lemon a lemon.
People have often been thought to understand concepts by the essential characteristics of a concept or ‘what a thing is.’ This is known as the classical theory of concepts, and it posits that a concept has necessary and sufficient conditions which determine its correct application. This is the conceptual basis for semantic normativity: a word, according to this theory, is basically a referential concept that has necessary and sufficient conditions for its correct usage. And it seems that when it comes to words and concepts, we think of them as having necessary and sufficient conditions that dictate whether something belongs in the category or not. So, is there any reason to believe that the family resemblance concept is actually how we know how to use a word correctly?
Overextension, Underextension, and Prototypes
One way to see if the family resemblance concept applies to how we learn and use words is to look at the early childhood development of language. When children first learn new words, they tend to overextend and underextend words. For instance, a child might only refer to their own dog as ‘dog’ when first using the word. This example would be an underextension because the child does not refer to all things within the ‘dog’ category as ‘dog.’ As our original example illustrated, an overextension would be if the child calls a cat a ‘dog’ because they see that a cat is also four-legged and furry. It is understandable why children would make these over- and underextension mistakes in the learning process, as the context in which they learn certain words might indicate various salient features about the referent that they regard with different degrees of importance. When referring to objects, children will overextend words because they do not have the appropriate words to describe the thing they are trying to talk about, so they find words with overlapping attributes to the given referent to use instead. If a child is trying to refer to a bear that she sees, she could call it a ‘dog’ because the attributes of ‘dog’ that the child has observed in previous uses of the word overlap with some of the characteristics she notices about the bear (fur, four legs). The child is using the shared attributes between bears and dogs to create a group for things that are furry, walk on four legs, and can run. And then the child associates this group of four-legged, furry, running things with the word ‘dog.’
When asked to point to a picture of a ‘dog,’ children will choose the photo of a dog first. When they are asked to continue to point to more pictures of ‘dogs,’ children will also point to pictures of things they had previously overextended the word ‘dog’ to (e.g. bears, cats). But they will never point to non-exemplars of ‘dog’ (e.g. train, house, anything that bears no resemblance to a dog). This indicates that they understand certain examples to be more prototypic than others: “they [the children] tend to choose objects that are prototypic insofar as the concepts underlying their use of the word is concerned.” For instance, the dogs in the neighborhood become the prototype of the word ‘dog’ because their parents will point and refer to the animal as a ‘dog.’ But anything with the observable characteristics of the dogs their parents pointed out to them—noisy, furry, four legs like a bear or cat—could also be referred to as ‘dog’ in the child’s mind. This is because they seem to learn words by collecting information about the characteristics of each referent of a word and then applying that word to novel referents based on shared characteristics with previous referents of the word. Through these associations, they begin understanding how objects are related to each other and begin forming categories based on these associations. Words are then connected with certain categories of associated attributes, and this is how the family resemblance model functions.
This also goes to show that children are not necessarily aware of the salient features of ‘dogs’ that we as adults use to identify a dog, since children can think that the word ‘dog’ also refers to cats and bears. But to adults, cats and bears have features that differentiate them from dogs. This means that when an adult points out a dog, the child does not necessarily know what features of the dog differentiate that dog from other similar things (e.g. cats, bears). Even though adults know the difference between dogs and cats, they are not much different from children when it comes to knowing what exactly a speaker means when making an utterance. Because we can never get inside someone’s head to know exactly what they mean (i.e. what salient features of the referent they have in mind) when they use a word.
The research on prototypes by Eleanor Rosch and Carolyn Mervis provides some evidence that the prototype/family resemblance theory applies to how adults conceptualize words too. Through six different experiments, they found a reason to think that prototypes of a given category are the ones with “the most attributes in common with other members of the category and least attributes in common with other categories.” The experiments showed that prototypes are not the members of a category that share a characteristic with all other members: “the salient attribute structure of these categories tended to reside, not in criterial features common to all members of the category which distinguished those members from all others, but in a large number of attributes true of some, but not all, category members.” For instance, a car would be the prototype of the vehicle category because it shares the most features with other members of the category (plane, train, bike), not because it has certain features that are common to all members of the vehicle category. These findings point toward an understanding of the meaning of words as categories that have overlapping similarities between all the members of the category. Because the prototypes of categories do not have essential characteristics as the classical theory of concepts suggests but rather have the most overlapping features with other members of the category as the family resemblance model suggests. The various studies on prototypicality within categories support Wittgenstein’s claim that the way we use words is not by analyzing whether a referent meets the necessary and sufficient conditions of the word referring to it. Instead, we seem to learn and use words through a string of connections between different uses of that word, some uses of a word being more prototypical than others.
We learn words through various examples in early childhood, and then continually refine our understanding of that word by observing other people’s use of that word. Our meaning of a word is thereby shaped by the examples we have learned and negotiated through social interactions with other people. And because of the similarities in the psychological, phenomenological, and social context within which most language users learn words, our understanding of word meanings converge enough to communicate fairly successfully. To make rules around the correct and incorrect usage of those words would be virtually impossible though. We all learn and use words through a vast array of different contexts that are similar enough for successful communication but different enough that we cannot know whether somebody else has the exact same idea of what any given words mean. The family resemblance model of semantics by no means explains all aspects of language acquisition and use, but it does explain how we might know what somebody is saying without actually knowing what they mean.
Bibliography
Kuczaj, Stan. “Thoughts on the Intensional Basis of Early Object Word Extension.” In The
Development of Word Meaning, edited by Stan Kuczaj and Martyn Barrett, 90-120. New
York: Springer-Verlag, 1986.
McNally, Thomas. Wittgenstein and the Philosophy of Language. Cambridge: Cambridge
University Press, 2017.
Putnam, Hillary. “Is Semantics Possible?” In Concepts: Core Readings, edited by Eric Margolis
and Stephen Laurence, 177-88. Cambridge: MIT Press, 1999.
Rosch, Eleanor and Carolyn B. Mervis. “Family Resemblances: Studies in the Internal Structure
of Categories.” Cognitive Psychology 7, no. 4 (October 1975): 573-605.
Wittgenstein, Ludwig. Philosophical Investigations. Translated by G. E. M. Anscombe, P. M. S.
Hacker, and Joachim Schulte. Rev. 4th ed. Chichester: Wiley, 2009.