Monthly Archives: April 2014

Beautiful Distinctions: Psychological and Phenomenal States

Aspiring PI

David Chalmers changed the frontier of philosophy of mind by resuscitating a key distinction between the phenomenal and psychological uses of mental terms. The former describing subjective qualitative aspects of mental phenomena and the latter describing physico-chemical and broadly neurological states which admit objective description. Better, he claims that these distinctions exhaust the landscape of mental phenomena. He draws one conclusion as follows:

“[O]nce we have fixed the psychological, phenomenal and relational properties of an individual, there seems to be nothing mental that can be independently varied. We cannot even imaginesomeone identical to me in the three respects mentioned above but who believes something different. There is simply not enough room in the conceptual space for the possibility. Intentional properties are in some ways less primitive than psychological and phenomenal properties in that they cannot be varied independently of the latter.” – David Chalmers, The Conscious Mind.

If such…

View original post 2,434 more words

Indeterminacy of Translation and Innate Concepts short version



            Throughout their careers, both Chomsky and Quine have dealt with Underdetermination[1] in different ways. Whenever Chomsky encounters UD he seeks to overcome it by postulating innate constraints; in contrast, Quine treats UD as a fact of life which simply has to be lived with. In this paper I will discuss the Indeterminacy of Translation[2]. Chomsky argues that the IDT amounts to nothing more than UD. He argues that the UD in physics is overcome by our innate science-forming faculty, and that UD in language raises no more difficulties than UD in physics. In the case of language, the IDT is overcome by innate constraints imposed by the rules of our language faculty.

Chomsky’s interpretation of the IDT has been accepted by the vast majority of contemporary cognitive scientists and they have proceeded to flesh out his proposal that the IDT can be overcome by innate constraints. However, when cognitive scientists are concerned with the IDT, they are typically interested in only one area of it, the inscrutability of reference. They typically argue that the inscrutability of reference is a form of UD in concept acquisition and that this UD can be overcome by postulating innate concepts which we use to learn our first words.  In this paper I will consider the attempt by contemporary cognitive scientists to overcome the inscrutability of reference by postulating innate concepts. Furthermore, I will analyse what effect the supposed overcoming of the inscrutability of reference by postulating innate concepts has on the indeterminacy of translation argument.

Most contemporary cognitive scientists follow Chomsky in arguing that the IDT amounts to nothing more than UD and that this UD can be solved by postulating innate concepts. In particular they are interested in problems of inscrutability of reference and argue that the fact that children do not fall foul of the inscrutability of reference can be explained by the fact that children are born with innate concepts.

In his book The Language Instinct, Stephen Pinker presents the inscrutability of reference as a form of UD:

A rabbit scurries by, and a native shouts, ‘Gavagai’ What does gavagai mean? Logically speaking, it needn’t be ‘Rabbit’. It could refer to that particular rabbit (Flopsy, for example). It could mean any furry thing, any mammal, or any member of that species of rabbit (say, Oryctolagus), or any other member of that variety of that species (say, chinchilla rabbit). It could mean scurrying rabbit, scurrying thing, rabbit plus the ground it scurries upon, or scurrying in general. It could mean footprint-maker, or habitat for rabbit fleas. It could mean the top half of a rabbit, or rabbit meat on the hoof, or possessor of at least one rabbit’s foot. It could mean anything that is either a rabbit or a Buick. It could mean collection of undetached rabbit parts, or ‘Lo Rabbit hood again, or ‘It rabbiteth’, analogous to ‘it raineth’. The problem is the same when the child is the linguist and the parents are the natives…Figuring out which word to attach to which concept is the Gavagai problem, and if infants start out with concepts corresponding to the kind of meanings that languages use, the problem is partly solved.(1994, 156)


Pinker then goes on to cite the research of the psychologist Ellen Markman to show that it is indeed the case that children are born with concepts which constrain the types of meanings that words can be given. It is Pinker’s view that Quine’s problem is a problem of UD which is solved by the fact that the concepts that children are born with will place innate constraints on the type of meanings that children can attach to words.

The child psychologist Paul Bloom draws a similar conclusion to Pinker:

These problems of reference and generalization are solved so easily by children and adults that it takes philosophers like Quine and Goodman to even notice that they exist. If we see someone point to a rabbit and say ‘gavagai’, it is entirely natural to assume that this is an act of naming and that the word refers to the rabbit and should be extended to other rabbits. It would be mad to think that the word refers to undetached rabbit parts or rabbits plus the Eiffel Tower. But the naturalness of the rabbit hypothesis and the madness of the alternatives is not a logical necessity; it is instead the result of how the human mind works. (2000, 5)[3].


The Harvard linguist Cedric Boeckx echoes the claims of the above thinkers:

Yet, if you think about it, the Gavagai situation is the one we all faced as children trying to acquire the meaning of words. How did we guess that elephant refers to that big grey animal with a long trunk? Because someone pointed to the animal and said elephant? But how did you know what exactly was being pointed at? Surely the finger couldn’t point at the whole elephant; it was your cognitive bias that interpreted the act of pointing in that way. (2010, 41)


So the above quotes clearly indicate that, within the realm of cognitive science, it is common-place to follow Chomsky and view the inscrutability of reference as a problem of UD, and indeed as a problem which can be solved by postulating innate concepts.  Pinker, Bloom and Boeckx are speaking only of the inscrutability of reference as it affects the child learning his first words. Chomsky goes further: he is arguing that the IDT is a form of UD which can be overcome by postulating innate concepts. He does not explicitly demonstrate how the IDT is overcome by innate concepts; rather, he merely points out that the situation in physics raises the same problems as the IDT problem. In the case of physics the problem is overcome by a science forming faculty, whereas in the case of the IDT the problem is overcome by our language faculty. His reasoning is that is that our innate concepts will ensure that when we are acquiring our first language, we will not think that a word like ‘gavagai’ refers to undetached rabbit part, etc. Rather, we will discover that the word merely labels our innately known concept of ‘rabbit’. The fact that our innate concepts will determine that our words have fixed meanings means that the IDT is incorrect in claiming that there are no facts about translation.

It could be argued that by postulating such innate concepts, these thinkers are merely begging the question against Quine by assuming the very determinacy of meaning which he is arguing against. However, Pinker et al. argue that they are not merely postulating innate concepts in order to overcome indeterminacy. Rather, they claim that there is independent evidence which supports their belief in innate concepts, and this evidence offers reasons to believe that UD will not be something that affects a young child who is learning his first language.



Empirical research in psychology indicates that children as young as four months already display conceptual knowledge. Jean Piaget’s experimental research led him to believe that young children at the age of five months do not have an adequate knowledge of objects and that they have to pass through a variety of developmental stages before they will acquire the full concept of an object. Piaget noted that children playing with a ball will not search for it if it runs out of their field of vision. He drew the conclusion that they do not have knowledge of object permanence at this age. As early as the seventies, psychologists such as Bower were questioning Piaget’s view by claiming that these babies do not search for the missing objects because of a problem of coordinating their movement.  And since the seventies, it has been common-place to try to test children’s concept of an object by using experiments that do not require coordinated sequences of actions. Bower’s studies have made surprising discoveries: (Here I am citing from Baillargeon, R., Spelke, E, and Wasserman’s paper ‘‘Object permanence in infants’’):

Bower’s studies have yielded four findings that seem to have provided evidence for object permanence in infants well below 9 months. First, 7 week old Infants were found to discriminate between disappearances that signalled the continued existence of an object (e.g. Gradual Occlusion), and disappearances that did not (e.g. gradual dissolution or sudden implosion). Second, 2 month old infants were found to anticipate the reappearance of an object that stopped behind the screen, ‘looking to that half of the movement path the object would have reached had it not stopped’ Third, five month old infants were found to show disruptions in their tracking when an object was altered while passing behind the screen: they tended to look back at the screen, as though in search of the original object. Finally, 5-month-old-infants were found to reach for an object that had been hidden by darkening the room. (1985, 195)


However, Bower’s experiments are not conclusive evidence that children are born with concepts of object permanence. Piaget himself claimed in a personal correspondence with Bower that he could interpret some of these experiments in a way which was consistent with his own theory (ibid., 97).  Because of the inconclusiveness of the debate between Bower and Piaget, Baillargeon et al. set up more rigorous tests to show that humans are born with concepts of object permanence. Quine, like Piaget, also argued that Bower’s experiments did not necessarily show that a 4-month-old child has a concept of an object:

True, an infant is observed to expect a steadily moving object to reappear after it passes behind a screen; but all this happens within the specious present, and reflects rather the expectation of continuity of a present feature than the reification of an intermittently absent object. (1969, 24)


Here Quine is interpreting Bower’s famous experiment to mean not that the young child has a concept of an object, but as indicating that the young child expects continuity in his experiences. Baillargeon et al.’s experiment aimed to test this view of Quine and Piaget’s.

They began by testing whether children have what they called The SolidityPrinciple, which was the principle that a solid object cannot pass through another solid object. An experiment was set up where an object was presented to the child and then occluded. Then another object was rolled towards the occluded object. If the child exhibited surprise that the object moved through the space which seemed to be occupied by first the object, then this would exhibit two things: first, that the child had grasped the Solidity Principle; second, that the child understood object permanence, because he knew the objects still existed when occluded. Baillargeon et al performed this experiment on four month old children. When completed the experiment showed that four-month-old infants do seem to reason according to the principle of solidity, and the principle of object permanence.

Quine explained Bower’s experimental results by saying that the child expected the object to return out the other side of the screen because of an expectation of continuity. The child saw the object-shaped thing move towards and behind a screen. So he expected this object-shaped thing to continue moving in the same direction as it started from; hence, the surprise when it did not move out the other side of the screen. For Quine this in no way indicates that our child has a sophisticated concept of an object. However, expectation of continuity does not explain what is going on in Baillargeon et al.’s experiment. The child is exhibiting two beliefs: first, the belief that object-shaped things do not disappear when occluded; second, that object-shaped things are solid, and hence cannot be moved through. Hence the surprise the child exhibits when the object appears on the other side of the screen as though it had passed through the original object.  Obviously, expectation of continuity does not explain what is going on in Baillargeon et al’s experiment. So, to account for both Bower and Baillargeon’s experimental results, Quine needs to postulate at a minimum that 4 month old children expect object-shaped things not to disappear when occluded, to move in undeviating paths unless stopped by another object-shaped blob, and to have solidity.  A child with knowledge of this sort seems to be clearly demonstrating a complex concept of an object, and a theory which, contrary to what Quine believes, precedes the learning of the syntax of quantification by about 2 years.

Quine argues that ‘Gavagai’ considered as an observation sentence has a clear stimulus meaning and can be correlated with non-verbal stimulation. The same situation occurs when the child says ‘Mama’; this is an observation sentence which has a clear stimulus meaning. According to Quine, the word is correlated with a scattered portion of what is going on in the child’s environment when he speaks. For Quine, the difficulty arises when we try to break down observation sentences into terms and predicates etc. When we tried to break down ‘Gavagai’, the observation sentence, into ‘gavagai’, the term, our problems begin. Our breaking down the observation sentence involves us using analytical hypotheses which go far beyond the empirical data. Quine notes that no behavioural evidence can decide whether ‘gavagai’ refers to ‘undetached rabbit part’, ‘particular instance of universal rabbithood’, ‘rabbit fusion’ etc. If we want to translate the words of the native we must unjustifiably impose our ontology onto the natives. The situation of the child is similar, though there are some notable differences. When the linguist is trying to translate terms like ‘gavagai’ he ends up imposing his own ontology onto the native; however, according to Quine, the child has no ontology. So when the child tries to move from using ‘Mama’ as an observation sentence to ‘mama’ as a term he will have little to go on. The child cannot impose his own ontology because he supposedly has none. As the child begins to gradually learn the apparatus of quantification and syntax, through induction, reinforcement etc, then he will gradually acquire the language of his peers. However, if Baillargeon et al.’s experiment is correct, then when children are learning to speak, their innate concept of object will determine that the UD facing the child learning his first language will be massively reduced.

There is also experimental evidence of children having theories of mind, as well as the theory of objects which Baillargeon et al discovered. These empirical discoveries indicate that prior to learning their first language, the child has an ontology consisting of objects, agents, and causality etc. which will partly determine how the child interprets the speech of the adults they engage with. So it is typically assumed that the child’s innate concepts of objects and agency will massively limit the UD facing the child as he learns his first language. However, while suggestive, this supposed evidence for innate concepts is derived from the study of pre-linguistic children so it does not directly deal with issues of the reference of words. In the next section I will discuss an experiment which was done on children who are in the early stages of learning their language. The experiment was designed to test whether Quine’s claim that children only develop an ontology after they have mastered the syntax of quantification is true.


The experimental research that I outlined above purported to demonstrate that children have innate concepts of object and agency. If the evidence is taken at face value, it implies that these innate concepts will ensure that the UD facing the child learning his first words will be decreased.  However, this argument runs contrary to Quine’s claim that we are only justified in attributing an ontology to children after they have mastered the syntax of quantification.  In this section, I will consider an experiment constructed by Soja, Carey, and Spelke which aims to test Quine’s claim.

In their 1990 paper ‘‘Ontological categories guide young children’s inductions of word meaning’’[4]  Soja, Carey, and Spelke tested whether the ontological distinction between objects and non-solid substances conditions the projection of word meanings prior to the child’s mastery of count/mass syntax. Quine denied that children make any ontological commitments prior to learning the syntax of quantification which helps them master the count/mass noun distinction. Here Quine is making an empirical claim that prior to grasping the count/mass distinction, an agent like Mother, a property like Red, and a non-solid substance like Water are on a par. According to Quine, it is only when a child has mastered the apparatus of divided reference through grasping the syntax of quantification that the child can distinguish these substances. Soja et al. set out to test these empirical claims of Quine’s. Before outlining their experiment I will outline Quine’s views on language acquisition.

                                             QUINE’S POSITION


Quine claims that when young children mouth words such as ‘Mama’, ‘Water’, or ‘Red’, we are in no position to state that they are using the words as terms which refer to the same things which we refer to by the sounds.

For though we may fully satisfy ourselves that the child has learned the trick of using the utterances ‘mama’, and ‘water’ strictly in the appropriate presences, or as a means of inducing the appropriate presences, we still have no right to construe these utterances in the child’s mouth as terms, at first, for things or substances. (1969, 7)


According to Quine, from our own mature perspective, we have come to view the child’s mother as a body which revisits the child from time to time, and water as a scattered object. However, from a behavioural perspective, we have little justification for imputing this ontology onto the child.  After making this negative point about our lack of justification for imputing our mature ontology onto the child, he then goes on to make a positive point about the nature of the child’s ontology.

But the mother, red, and water are for the infant all of a type: each is just a history of sporadic encounter, a scattered portion of what goes on. His first learning of the three words is uniformly a matter of learning how much what goes on about him counts as the mother, or as red, or as water. It is not for the child to say in the first case ‘Hello mama again’, in the second case ‘Hello another red thing’ and in the third place ‘Hello more water’. They are all on a par: Hello more mama, more red, and more water. (ibid., 7)


Here Quine is clearly claiming that young children who use words, such as ‘Mama’, ‘Red’ and ‘Water’ are not distinguishing them in terms of being respectively Objects, Properties, and Non-solid substances. His reason for arguing so is that we have no positive behavioural evidence to support the claim that children make such distinctions, and in the absence of such positive evidence, there is little reason to impute such a rich ontology to young children. Quine’s claim is that we should only attribute to children the ability to distinguish between Objects, Substances and Properties when we have behavioural evidence which supports us making this distinction.

Progressively, however, the child is seen to evolve a pattern of verbal behaviour that finally comes to copy ours too closely for there to be any sense in questioning the general sameness of conceptual scheme. For perspective on our own objectifying apparatus we may consider what steps of development make the difference between ‘mama’-babbling infant who cannot be said to be using terms for objects, and the older child who can. It is only when the child has got on to the full and proper use of individuative terms like ‘apple’ that he can properly be said to have taken to using terms, and speaking of objects. Words like ‘apple’ are not words like ‘mama’ or ‘water’ or ‘red’ are terms whose ontological involvement runs deep. To learn ‘apple’ it is not enough to have learned how much of what goes on around you counts as apple; we must learn how much counts as an apple, and how much as another. Such terms possess built in modes of individuation. (ibid., 8)


Now Quine acknowledges that the child may learn ‘apple’ in the same way that he learns ‘mama’ or ‘red’ but he goes on to say that the child will never master ‘apple’ in its individuative use until he gets on with the scheme of enduring physical objects. And in order to get on with the scheme of enduring physical objects, the child will need to master the apparatus of identity, difference etc.  Quine claims that to be able to tell if the child has got the trick of individuation down, we need the following:

How can we ever tell if the child has got the trick of individuation? Only by engaging him in sophisticated discourse of ‘that apple’, ‘not that apple’, ‘an apple’, ‘same apple’, ‘these apples’, ‘another apple’. It is only at this level that a palpable difference emerges between genuinely individuative use and the counterfeits lately imagined.(ibid., 9)


It is at this stage that Quine claims that we are justified in attributing an ontology to the child. Prior to that, attributing an ontology to the child is making an unsupported conjecture which is not justified by the facts.  For Quine, our child learns the adjectives ‘same’, ‘another’, ‘an’, ‘that’, ‘not that’ contextually. First the child gets used to various longer phrases which contain them, and he gradually develops appropriate habits in relation to the component words as common parts and residues of those longer forms. He further speculates that the contextual learning of all of these various different particles goes on simultaneously, so that we gradually adjust them to each other as a coherent pattern of usage is evolved (ibid., 10). So the story of child ontology as Quine tells it is that the child’s words just represent scattered portions of what goes on and do not distinguish between Objects, Properties and Substances.

So Quine’s picture of a child learning language and the ontology which is implicit in this language involves pared down assumptions according to which we attribute to the child no more than is necessary to explain his verbal behaviour. Quine treats the babbling which a child begins to emit at the age of 12 months as a form of operant behaviour which is omitted rather than elicited. He claims that the family of the child will reinforce the child’s verbal behaviour (such reinforcement made possible by the child’s pre-linguistic quality space) in such a manner that the child’s use of observation sentences such as ‘mama’ will reliably distinguish between ‘mama’ portions of the environment, and ‘non-mama’ portions. However, at this stage we cannot credit the child with having an ontology; from the point of view of external verbal behaviour we have no reason to attribute to the child a concept of ‘Mama’ as a name of a spatio-temporal object, as opposed to being a name of a mere mass term like ‘Water’.  It is only when we engage the child in discourse and he can answer the questions using terms such as ‘not that mama’, ‘same mama’, ‘another mama’ etc., that we are justified in attributing to the child a concept of ‘mama’ as an object as opposed to scattered portion of mama environment etc.

                                    TESTING QUINE’S CLAIMS

Soja et al. conducted their experiments to test Quine’s claim (1960, 1969) that young children only develop an ontology after they have grasped the syntax of quantification.  Contrary to Quine, they claimed that young children have a distinction between different ontological categories prior to grasping the syntax of quantification, and that in fact these ontological categories constrain the process of language learning. They distinguished their views from Quine’s in the following way:

According to Quine, then, when children hear a new word, the meaning they assign to it is determined by procedure 0:

Procedure 0: Conclude that the word refers to aspects of the world that share salient properties of the perceptual experience when the word is used. (1991, 182)


Soja et al. proposed a different view of the procedures children use when they learn a new word; their procedure assumed that the child had ontological categories prior to learning the syntax of quantification.

Procedure 1    Step 1: Test to see if the speaker could be talking about a solid object; if yes,

Step 2: Conclude the word refers to individual whole objects of the same type as the referent.

Procedure 2 Step 1: Test to see if the speaker could be talking about a non-solid substance; if yes,

Step 2: Conclude the word refers to portions of substance of the same type as the referent.  (ibid., 183)


Soja et al. proposed an experiment which would decide between these two different proposals about how children learn new words.

One way for Soja et al. to test whether Quine was correct, or whether they were correct, was to test how children generalised when they learned words for different objects. If children could generalize prior to a grasp of the count mass syntax this would be evidence that Quine was wrong. They ensured that the experiment was done on children who are below the age of 2 ½, the age at which children master the syntax of quantification.  They tested how children generalise words to non-solid substances as well as to objects. If Quine is right that children generalise names by using Procedure 0, then children will generalize names based on shape whether the name originally refers to an object or a non-solid substance.


Twenty-four 2-year-olds from the Greater Boston area were recruited and randomly placed into two groups (informative syntax groups, and neutral syntax groups), with equal numbers of boys and girls in each group. Each testing session began with two familiar trials: one object trial and one substance trial. The stimuli in the familiar object trial were a blue plastic cup, a white Styrofoam cup and cup pieces. The stimuli in the familiar non-solid substance trial were peanut butter and Play-doh. These trials followed the same format as the unfamiliar trials described below.  The two familiar trials were followed by eight unfamiliar trials: four object trials and four substance trials which were intermingled. The subjects were tested on each trial on two separate occasions. Eight novel words were used: ‘blicket’, ‘stad’, ‘mell’, ‘Coodle’, ‘doff’, ‘tanninn’, ‘fitch’, and ‘tulver’ (ibid., 187)


The test involved presenting the child with an unfamiliar object, e.g. a plumbing T-shaped pipe, and giving the child a name for the object, e.g. blicket. In the neutral syntax condition the child is told ‘This is my blicket’. The experimenter then continued to talk about the object using ‘my’ ‘the’ and ‘this’ as determiners. She and the subject then manipulated the object. The object was then placed to the side and two other sets of objects were then presented to the subject.  One set consisted of objects of the same sort as the original but made of a different material, e.g. a plastic T shape; the other set consisted of objects of the same material but a different shape i.e. bits of metal.  The experimenter then said ‘Point to the blicket’


The child was shown an unfamiliar substance, and told ‘This is my stad’. The experimenters referred to the substance using only the determiners ‘my’ ‘the’ and ‘this’. The experimenter and the subject talked about the substance and played with it.  In the presentation of test substances, the subject was shown two substances, the original and the new one, and told ‘Point to the stad’.  The original substance was in the alternative configuration, whereas the new substance was in the configuration used originally with the named substance.  There were four pairs of substances: (1) Dippity-do (a setting gel), and lumpy Nivea (a hand cream mixed with gravel), (2) Coffee (freeze dried) and Orzo (a rice shaped pasta), (3) Sawdust and leather (cut to tiny pieces), (4) Crazy foam and Clay. Of each pair one member was named and the other was used as the alternative to the original in the test presentation. Each member served in both roles across subjects.


This condition differed from the neutral syntax condition only in the determiners and quantifiers used when naming the original stimulus. The experimenter introduced an object trial in the informative syntax condition with ‘This is a blicket’ and used ‘A blicket’ and ‘Another blicket’ in subsequent discussions. Substance trials in the informative syntax condition were introduced with ‘This is stad’ and in subsequent discussion the experimenter continued to omit determiners or use ‘some’ or ‘some more’. This was the only difference between the different informative and uninformative trials. In the familiar word trial subjects differentiated the object and the substance trials as predicted.

                                   WORD LEARNING TRIALS

Subjects differentiated the two types of trials.  Responses were consistent with shape and number on the object trials, and were not consistent with shape and number in the substance trials.

                                     WHAT THE TEST SHOWS

If before the child has grasped the syntax of quantification the child differentiates in the above manner, this shows that the child is not generalizing the word-based perceptual similarity, but is doing so based on the type of object he is presented with. So, for example, if he was generalizing according to an innate perceptual similarity quality space which focuses on shape then why does this not work for substances? The answer is because the child recognises that objects and substances are distinct ontological categories. Soja et al. summed up their results as follows:

In sum, the children chose according to object type when the stimulus was an object and according to substance type when the stimulus was a non-solid substance. There was no effect of the syntactic context: performance was neither facilitated nor hindered by the additional syntactic information.

The data from Experiment 1 show that different inferences about the meaning of a newly heard word are drawn according to the ontological status of its referent. If the word refers to an object, the child’s projection respects shape and number, and ignores texture, color, and substance. If the word refers to a non-solid substance, the child’s projection ignores shape and number, respecting texture, color and substance. (ibid., 192)


From this experiment, Soja, et al. claim to have shown that Quine’s view of how children learn language is incorrect because the experiment shows that, contrary to what Quine claims; children do indeed have a distinction between different ontological categories prior to grasping count/mass syntax. It also shows that these innate ontological categories are what help a child learn a language and not the apparatus of quantification. Soja et al.’s experiment purports to have shown that children learn words according to ontological distinctions which they exhibit knowledge of prior to learning a language.  So their experiment, along with the experiment of Baillargeon et al., purports to show that Quine is wrong in his story about how children develop their concept of an object.

Again, the relevance of this experiment to the IDT should be apparent. While Soja et al. purport to have shown that children have an ontology prior to learning their first language, Baillargeon et al. purport to have shown that young children interpret the meaning of new words using this pre-linguistic ontology. Quine claims that linguists construct their analytical hypotheses by imposing their ontology onto natives. These two experiments try to demonstrate that a child interprets the sounds of his peers by imposing his own innate concepts. So according to this view, there are facts about meaning which are imposed by our universally shared concepts.


Research by Baillargeon et al. which I discussed in Section 1 Part 3 indicates that four-month-old children seem to have implicit knowledge of the following five principles about objects:

(1) The principle of solidity.

(2) The principle that objects are three-dimensional.

(3) Objects tend to move on undeviating paths.

(4) Objects move continuously through space and time.

(5) Objects only move when contacted by another object. (Children make exceptions to this principle when they are viewing intentional objects)


This work of Baillargeon et al. indicates that children from a very young age have an intricate knowledge of objects; on the other side of the coin the work of people like Simon Baron-Cohen[5] indicates that prior to learning language, normal children view other people as intentional agents with beliefs, desires etc.  Furthermore, there is evidence that such children who lack this theory of mind suffer severe language learning difficulties. The evidence for a pre-linguistic theory of mind has been summed up succinctly by the psychologist Paul Bloom:

What understanding do prelinguistic children have about the minds of others? Consider first sensitivity to what other people are attending to. By around nine months, a baby will naturally follow its mother’s line of regard (Butterworth, 1991), and will also follow her pointing gestures (Murphy and Messer 1977), at about the same age, babies can monitor their parents’ emotional reactions to potentially dangerous situations and react accordingly. For instance when seeing a spider, a baby will be less likely to approach it, if its mother seems fearful than if she seems happy (Zarbatany and Lamb 1985) and when babies are uncertain or hesitant, they check what their mother is looking at and how she is reacting (Bretherton, 1992)… (2000, 67)


There is, then, some compelling evidence from both a logical and an empirical point of view which seems to indicate that children have pre-linguistic concepts which they use to help them learn their language. Furthermore, it is clear from the empirical studies that children exhibit knowledge of intentional agents and object constancy prior to their learning their first words at the age of 12months.

This evidence so far indicates the following: prior to the age of 12 months when young children begin to learn their first words, they exhibit a grasp of concepts of agency, object, substance etc. Young children also do not need to grasp the syntax of quantification to have ontological commitments; in fact there is evidence that young children (below the age of 2 ½) have an ontology which constrains the meanings that they can give to words. This evidence is typically used to argue that the inscrutability is overcome by our innate concepts.


Markman begins her book Categorisation and Naming in Children: Problems of Induction (1989) by discussing Quine’s IDT problem and claims that the young child is faced with the same problem as the adult linguist. Her empirical research is conducted to the end of seeing how the child does in fact overcome the problem. She begins by discussing the research of Piaget on young children, which indicates that young children categorise objects primarily in terms of thematic relations as opposed to taxonomic relations (e.g. a spider and a web are categorised together because they are both part of the same theme: the spider building a web).

The fact that children prefer to organise their experiences according to thematic relations poses a serious problem for theorists who claim that innate constraints on interpretations will limit the problem of IDT. Markman summed up the problem as follows:

Children seem to readily learn terms that refer to object categories. Their vocabulary is filled with words such as ‘ball’ and ‘dog’, simple concrete nouns referring to object categories. Yet children often notice and remember thematic relations between objects more readily than an objects category. How is it that children readily learn labels for categories of objects if they are attending to relations to objects instead? To take a concrete example, imagine a mother pointing to a baby and saying, ‘baby’. Based on the classification studies, we should assume that the child will be attending to the baby sucking on the bottle, or to the baby being diapered. Why, then, doesn’t the child infer that ‘baby’ means something like ‘baby and its bottle’ or ‘baby and its diaper’?(1989, 27)


Therefore, if we accept the results of the categorisation tests, and if we further accept that the IDT is overcome by pre-linguistic biases, then it seems to follow that children will think that words like ‘baby’ will pick out thematic events like ‘baby sucking on his bottle’, and this view runs counter to our intuition of what words typically mean.

Markman proposed a solution to this problem. She claimed that when children are interpreting new sounds, they will typically assume that the new sound refers to a whole object. Markman then constructed a series of experiments which were designed to test this proposal. Markman’s proposal is as follows:

Children may be biased to treat novel labels as referring to whole objects (the whole object assumption) and to treat them as referring to objects of the same type (the taxonomic assumption). (ibid p, 27)


Upon testing these assumptions the experimental evidence verified that these assumptions were indeed correct.

                                         EXPERIMENTAL EVIDENCE

Markman’s experiment was done on children between the age of two and three, and was designed to test whether children exhibit what she called the taxonomic and whole object assumption.  Her prediction was that children should interpret novel labels as labels for objects of the same type, as opposed to objects that are thematically related. And to test this prediction, she organised a test which compared the way children organised objects when they were given a novel object label and the way they organised objects when they were given no label. There were two conditions in the experiment: a no word condition, and a word condition. In the two conditions the children were shown a target picture; they were then shown two other pictures and asked to select one of them as being the same as the target (ibid., 27).


The children were introduced to a hand puppet and told to put the picture they choose into the puppets mouth. The child was then shown a picture such as a picture of a dog, and then told ‘Look carefully. See this? Find one that is the same as this.’


This condition was identical to the no word condition with one exception. In this condition the child was told that the puppet could talk in puppet talk. They were told to listen very carefully to find the right picture. The puppet gave the target picture an unfamiliar name and used the same name in the instructions for picking a choice picture. For example, the puppet might say, ‘See this? It is a sud. Find another sud that is the same as this sud.’(ibid., 28)

The results of the experiments in the two different conditions are instructive. In the no word condition, the child who had to choose between another category member and a thematically related object, often chose the thematically related object. They selected other category members a mean of only 59% of the time, which is no better than chance. In the novel word condition, the children selected categorically the vast majority of the time. They chose the other category member a mean of 83% of the time, which is better than chance.

So these results supported Markman’s prediction that when young children are learning the meaning of a new word they switch to taxonomic understanding of the meaning of the word. In other words, the child will assume that ‘Gavagai’ refers to rabbit as opposed to rabbit running in the grass.

However, this experiment is not directly relevant to the claims which Quine is making in Word and Object. When Quine is talking about the inscrutability of reference he typically makes the following point:

Point to a rabbit and you have pointed to a stage of a rabbit, to an integral part of a rabbit, to the rabbit fusion and to where rabbit-hood is manifested. Point to an integral part of a rabbit and you have pointed again to the remaining four sorts of things; and so on around. Nothing not distinguished in stimulus meaning itself is to be distinguished by pointing unless the pointing is accompanied by questions of identity and diversity.(1960, 53)


Markman’s experiment indicates that children will make a whole object assumption in the new word condition 83% of the time. The relevance of this fact for Quine’s concerns is questionable. If a child interprets new words using the taxonomic assumption instead of the thematic assumption, then this is an interesting fact about child psychology but it does not refute Quine’s argument. If a child does make the taxonomic assumption for novel words he hears and continues to make this assumption when he uses the word, this will not overcome Quine’s concerns. The child may not use ‘Gavagai’ to refer to rabbit running in the grass, however nothing in the experiment can tell us whether the child uses the term to refer to an undetached rabbit part or a particular instance of universal rabbithood. The point of Quine’s thought experiment is that we can have no behavioural evidence which can indicate which of these interpretations is the correct one. We can try and find out if we engage the children or native in questioning, for example, are you speaking about a rabbit or an undetached rabbit part? However, in the case of the native we can only ask them when we have learned their language and hence have made assumptions about reference which will be hidden from experience.  While in the case of child, we will only be able to ask them after they have learned our language and hence have adopted our ontology. So Markman’s experiment does not really touch the issues which concern Quine. In Markman’s experiment, it can be decided whether the child is making a thematic choice or a taxonomic choice depending on what objects they pick. However, Quine’s example of ‘gavagai’ cannot be decided in such a manner because every time the child picks out rabbit he will also be picking out an undetached rabbit part etc.


We have seen that a number of contemporary linguists and psychologists view the IDT as a form of UD which faces the child when he is trying to learn his first words. These theorists argue that innate constraints overcome this problem of UD. Merely postulating innate constraints to overcome the problem is question-begging, so the theorists need independent evidence to support their postulation of innate concepts. I have reviewed some different experiments which are typically cited as evidence to support the view that the UD facing the child is overcome because of innate constraints.

The first experiment indicates that children at the age of four months have a concept of what it is to be an object which includes beliefs that it is solid, three-dimensional, tends to move in undeviating paths, moves continuously through space and time, and only moves when contacted by another object. This complex concept of an object exists prior to a child learning language and is clear evidence that the child has ontological commitments prior to learning language and the syntax of quantification. The second experiment showed that before the age of two and a half when they have mastered the syntax of quantification, children could, contrary to what Quine claimed, distinguish between mass terms and count nouns. The experiment proved that when learning new words for solid substances children generalised based on shape and number, and when learning new words for non-solid substances generalised based on texture and colour. This experiment showed that children can distinguish between these substances prior to learning count nouns and mass terms, and in fact this pre-linguistic ontology helps the child learn the distinction between mass and count nouns, not vice versa.

The third experiment showed that when children are learning a new word they are constrained to make a taxonomic assumption when deciding what the word refers to. This experiment is interesting in showing that when children are learning a new word they seem to be constrained in how they interpret the word. However, the experiment does not deal with the concerns of Quine, because even if the taxonomic assumption is in place, this still  leaves the question open of whether ‘gavagai’ refers to undetached rabbit part, particular instance of universal rabbit-hood etc.

We saw above that thinkers like Pinker, Bloom and Boeckx follow Chomsky in thinking of the IDT as a form of UD which can be overcome by innate concepts. They typically cite the experimental work of Baillargeon, Spelke, Soja and Markman as evidence that children are born with the innate concepts which will ensure that children will overcome the problems raised by the IDT.  However I argue now that these experiments are of limited significance when it comes to evaluating Quine’s philosophical position.

When Quine speaks of ontological commitment only being possible when we have mastered the syntax of quantification he is primarily talking about an explicit worked out ontology. A child having implicit beliefs about the behaviour of objects is of limited philosophical significance. It is only when a creature begins to speak about this and that object, and to make claims which are true or false about objects, that we can really evaluate their ontic commitments. My primary concern is whether any of the three experiments offer evidence that the IDT viewed as UD is overcome by innate constraints.

The three experiments offer no evidence that the UD is overcome for children learning their first language because of innate constraints. Markman’s experiment gives us no reason to believe that a child can decide between ‘gavagai’ in the sense of rabbit as opposed to undetached rabbit part, particular instance of universal rabbithood, or rabbit fusion etc. So her experiment is not relevant to the questions which Quine is raising when it comes to the Inscrutability of reference. It is typically believed that the evidence provided by people such as Baillargeon et al. etc. is evidence of innate concepts, and, indeed, the authors of these papers typically argue in this manner. However, the fact is that they only provide evidence for pre-linguistic concepts, not for innate concepts. It could be argued that if children have pre-linguistic concepts which determine that they will interpret ‘rabbit’ as referring to rabbit then by claiming that the concept RABBIT is innate we have solved the inscrutability of reference for a child learning his first words.  His innate concept of RABBIT willdetermine that ‘rabbit’ will mean rabbit. Furthermore if our innate concepts determine what the words mean, then this will mean that our analytic hypotheses’ are fixed from the start.  However, the empirical evidence indicates that this argument does not work.

First, the experimental evidence which I reviewed above merely shows that children have pre-linguistic concepts; they do not demonstrate that children have innate concepts. If we assume that a child has a pre-linguistic concept of GAVAGAI, the question arises of how the child learns this concept. If the concept of GAVAGAI is supposed to refer to Rabbit as opposed to undetached rabbit part, particular instance of universal rabbithood, or rabbit fusion, then one is faced with the question of what evidence the child used to decide amongst these different meanings of the concept? Quine’s point was that there was no behavioural evidence which can decide between the various possible meanings of GAVAGAI. So if there is no evidence which the child can use to decide between the various different meanings of GAVAGAI, then Chomskians are faced with an APS for concept acquisition. If they want to argue for determinacy about the meaning of the concepts, they need to overcome the APS. A possible way to overcome the APS is to argue that children are born with innate concepts. However, to argue in this manner is to merely beg the question against Quine by assuming the very determinacy of meaning which is in question here.

To be clear, the issue here is not innateness per se, but innate concepts. Human intuitive physics differs from those of apes in some respects, and other creatures in many respects. The differences between these creatures are obviously to do with their innate genetic endowment. However, nativists typically claim that certain intuitive theories of object behaviour, the behaviour of agents etc. is written into the genetic code of most if not all humans. And they claim that these theories will grow in the mind if children are situated in any normal environment. Non-nativists claim that humans differ from other animals because they are innately wired up to be able to recognise statistical regularities in their environment which will help them learn the structure of the world they are born into. The important point to note here is that the experiments of Soja et al. and of Baillargeon et al. do not tell us whether children learn their theory of object behaviour statistically or because they are born with innate concepts of objects etc.  Quine argues that we learn our concepts through observation and trial and error. However he notes that this mode of learning will not determine what the meaning of GAVAGAI is. The experimenters who I reviewed above all argue that such concepts are innate; however, as we have seen, the evidence does not support their claims on this point. Without such evidence in place, they can simply state that innateness overcomes the UD but it remains just that, a statement, a question begging-assertion rather than an empirical fact.

It is also important to note that the types of experiments which they constructed are in principle blind to Quine’s concerns. Quine is claiming that from a behavioural point of view there is no fact of the matter as to whether ‘gavagai’ refers to undetached rabbit part, particular instance of universal rabbithood, rabbit fusion etc. None of the experiments above can tell what ‘gavagai’ refers to, nor can any behavioural evidence in principle, because, as Quine says, every time we point to one of them we are pointing to the others as well. So the experiments do not really affect Quine’s IDT argument.

The situation is this: one of the premises of the IDT, the inscrutability of reference, says that there is no fact of the matter as to whether ‘gavagai’ refers to undetached rabbit part, particular instance of universal rabbithood, it rabbiteth, etc. One can avoid this conclusion by saying that we know intuitively that there is a distinction between the various potential references of the term, and we know this because we are innately constrained to interpret ‘gavagai’ as referring to rabbit. However, this move merely begs the question against Quine.  Likewise it could be argued that Quine is begging the question against Chomsky et al. by assuming that there is no determinacy of meaning.

We have seen that if concepts are learned statistically through incremental data processing, then there can be no evidence to distinguish between the various interpretations of ‘gavagai’. If, on the other hand, we assume that children are born with innate concepts which distinguish between the various different interpretations, then our problem is supposedly solved. However, given that behavioural evidence cannot decide between the various different interpretations, the postulation of innate concepts that solve the problem amounts to merely assuming without evidence that the problem does not exist.


I have reviewed some empirical research on the nature of concepts which purports to show that Quine is incorrect about how children develop their ontology.  But nothing in this research tells us much about facts of translation; in fact it is only the strained analogy with the situation of the field linguist and the situation of the child that makes us think there is any connection at all. We have seen that thinkers such as Chomsky, Pinker, Markman, and Boeckx all write as though the two situations were identical. It is my contention, however, that this analogy should be dropped, that we should not compare the situation of the field linguist with the situation of the radical translator. Firstly, the field linguist differs from the child in that the field linguist has an explicit theory of the world which will affect how he interprets the sounds of the native, whereas the child supposedly operates using an implicit theory which will automatically map certain sounds onto certain interpretations.  The linguist will have a total network of beliefs about the world which he will bring to the translation situation. These beliefs will help him translate what the native is saying, but they will also ensure that we end up translating them according to our own lights. It could be argued that this situation is no different than the situation that the child finds himself in, and that the same solution is possible in terms of positing innate constraints on the interpretations the translator can make. Furthermore, it could be argued that these constraints are justified, given that the translator and the native are both members of the species homo-sapiens and so are subject to the same innate constraints. Such a move, I argue, is not justified by the best evidence that we have available to us at the moment. Firstly, as we have seen above, we do not as of yet have sufficient evidence that innate concepts exist, for all  that contemporary cognitive science has shown is that children have complex concepts from a very young age. Secondly, even if Homo-sapiens are, indeed, born with innate concepts, this would have limited significance for questions of translation.

Let us assume (wrongly) that contemporary cognitive science has proven that humans are born with innate concepts. So when our native begins to converse with the linguist, there will be a plethora of innate constraints which will in some ways determine how communication between the two takes place. These constraints have a lot in common with the types of constraints which Markman, Soja, etc talk about. Thus, one such constraint will be fact that the linguist will automatically track the eye movements of the native and try to triangulate with the natives on shared objects of experience. The linguist will use the language of the eyes to try out translations of the native’s utterances by mouthing ‘Gavagai’ while looking at the rabbit.  But obviously such innate constraints do not entirely determine what an adult is referring to when he uses the terms. An adult speaks about reality and communicates with others based on what he has learned about the world through his scientific education, religious education, social education, etc. When an ordinary adult speaks, he will speak in a language which reflects the ontological commitments of the community which he was socialised in. The discourse of any community will be shot through with assumptions which are justified in light of the total theory of the experts and the citizens of the community.  It is possible that a child may be born with a folk biology which determines that it sees a moving furry object with eyes as an agent.  And an adult may have the same innate bias wired in his brain; however, such a bias will not determine the speech of the adult. Imagine, for example, Putnam’s thought experiment that it has been discovered that rabbits are remote-controlled spies from Mars. In that situation, people may be innately biased to see rabbits as agents but would have scientific evidence that this bias was in fact false.  Various theories of the world which are learned through pain-staking trial and error will influence what people think they are referring to when they refer to certain entities.  When we are trying to understand the discourse of a native, we have no way of knowing what sort of ontological commitments they are adhering to. Do they have the same information about rabbits being remote-controlled spies as we do? The natives could be Platonists, or Nominalists, they could be Dualists or Monists. Without knowing their ontology prior to translation, we will have no choice but to assume that their ontology is the same as ours. This will mean, of course, that we are translating their language according to our own lights. The experiments of Markman, Soja and Spelke are not really relevant when it comes to facts about translation.  The native and the field linguist may both be members of the species Homo-sapiens and hence will presumably learn language subject to the same constraints as scientists like Markman, Soja et al. claim. Nevertheless, the fact that we develop sophisticated theories of the world which go far beyond the crude ontology of children means that understanding the constraints on language learning will give us little to go on in constructing a translation manual when we are trying to interpret our native.

Quine is quite explicit on this point: the IDT argument is separate from questions of how a child learns his first language. When Quine is speculating on how a child learns his first language, he is doing so because he believes that understanding how humans in fact develop their ability to refer will help us better understand the process of reference. And understanding our process of reference better will help us better grasp the process of ontic commitment.

We have seen in this paper, that despite the fact that most cognitive scientists believe that the IDT has been overcome by the postulation of a language faculty, the evidence indicates otherwise.  None of the experimental evidence demonstrates that children are born with innate concepts. So neither the Inscrutability of Reference nor the IDT is overcome by innate concepts. The fact is that, as of yet, we do not have enough evidence to indicate whether children are born with innate concepts. Furthermore, even if children are born with innate concepts, this fact would have no relevance for the IDT unless it could be shown that the innate concepts people are supposedly born with cannot be changed by experience.     



Baillargeon, R, Spelke, E and Wasserman, S. 1985 Object Permanence in Five Month Old Infants. Cognition 20 pg 191-208.

Baron Cohen, S., 1995 Mindblindness: An Essay on Autism and The Theory of Mind. Cambridge MA: MIT Press.

Bloom, P. 2000 How Children Learn the Meaning of Words. Cambridge MA:  MIT Press

Boeckx, C. 2010 Language in Cognition. London: Wiley-Blackwell. Boeckx, C. 2010 Language in Cognition. London: Wiley-Blackwell

Chomsky, N. 1969. “Quine’s Empirical Assumptions” Synthese 21 pp 53-68

Chomsky, N. 1980b. On Cognitive Structures and Their Development. In M Piattelli-Palmarini ed Language and Learning: The Debate between Jean Piaget and Noam Chomsky. London:  Routledge and Kegan Paul, pg 35-52

Markman, E. 1989 Categorization and Naming in Children. Cambridge MA: MIT Press

Piaget, J. 1926 The Language and Thought of the Child. London: Routledge

Pinker, S. 1994 The Language Instinct. London: Penguin Books.

Quine, W.V 1960 Word and Object. Cambridge MA: Harvard University Press.

Quine, W.V 1975  ‘‘Empirically Equivalent Systems of the World’’. Erkenntnis 9 pg313-28

Quine, W.V 1970 ‘‘On the Reasons for the Indeterminacy of Translation’’.  The Journal of Philosophy 67. Pg178-183

Soja, N., Carey, S., and Spelke, E,. 1991 ‘‘Ontological categories guide young children’s inductions of word meaning: Object terms and substance terms.’’ Cognition, 38, pg179-211


[1]Henceforth Underdetermination will be known as UD.

[2]Henceforth the Indeterminacy of Translation will be referred to as IDT.

[3] While Pinker’s and Bloom’s arguments on this narrow point are similar, obviously this does not imply that they hold the exact same view on the nature of the mind.


[5]See Simon Baron Cohen:Mindblindness: An Essay on Autism and Theory of Mind.