Language as grist to the mill of cognition

Systematicity of thought concerns the empirical fact that the ability to entertain certain thoughts is pertinently linked to the ability to entertain certain others. For instance, if S can infer P from P&Q then S can also infer Q from P&Q and P from P&Q&? See Fodor and Pylyshyn (1988: 46–48). Productivity of thought concerns the fact that we can entertain infinite many thoughts on the basis of presumably finite means.

Note that this does not mean that such animals have no mental life at all, nor that we cannot usefully use mental concepts in explaining and predicting their behaviour. Rather it means that the extent to which we can think of such creatures as having attitudes and a mental life is measured by the extent to which we can assign determinate propositional content to the attitudes we would ascribe to them (Malpas 2009). I do not further elaborate on Davidson’s views here.

Most of these views are reported in Clark (1998). See also Sutton (2002) for a discussion.

Even though I am not pressing on this point here, the Whorfian hypothesis could be seen as orthogonal to what Patterson and Fushimi (2006) report. In particular, they studied speakers of fundamentally different languages (English, where written words represent the sounds of words, and Japanese, where written words represent the meanings of the words) with the same causes and neuroanatomical locations of brain damage and showed that the brain's organisation of language is in fact the same regardless of the language the subject speaks.

Hebbian rule of learning implemented by Long-Term Potentiation (cf. Martinez et al. 2002; Shors and Matzel 1997).

I take sub-activated representations to be representations that are not fully blown in consciousness but which allow enough current to go through the neurons grounding their connections. In turn, this influences the conceptual network’s activation pattern. See Trafford and Tillas (2015); Elman et al. (1996) for detailed discussions.

These were fMRI studies.

Further evidence in support of the suggested associationistic view of thinking can be found, amongst others, in the work of Elman et al. (1996), who argue that artificial neural networks can be highly constrained by the network’s current weight assignment. More specifically, the pattern of activation set by a connectionist network is determined by the weights, of connections between the units. Given that these weights mirror the effects of the synapses between different neurons, different levels of activation of synapses place a significant constraint on which ideas the mind can explore next. Thus, sub-activation of certain neuronal ensembles constrains or drives activation in a specific way, i.e. towards specific thoughts associated with the sub-activated links of the linguistic network in question.

As shown below, the same problem lies at the heart of the problem of compositionality of thought.

Structure and content are different since there could be mental atoms that have propositional content.

This is the notorious challenge Berkeley (1710/1957) posed to Locke (1690/1975).

See Tillas 2010; 2015 for an acquisition account concepts including widely heterogeneously instantiated concepts, like fuel and danger, as well as lofty concepts that do not pick out tangible entities, like justice and democracy.

Similar ideas can be found amongst others in Aristotle (1961) and Berkeley (1710/1957).

Perhaps an intention to avoid similar problems is what lies behind Locke’s motivation to establish an abstraction model of how general ideas are formed. See Tillas 2014 for a detailed discussion of related issues.

See also Bechtel and Abrahamsen (1992) who describe connectionism as ‘associationism with an intelligent face.’ (p. 103).

Prototype theory builds upon associationism to the extent that prototypes are reducible to a weighted set of features or similarities, and that formation of representations (in prototype theory) is normally seen as following Hume’s raw similarity based model (Ryder and Favorov 2001).

On these grounds, Fodor and Lepore (1996) argues that prototypes cannot combine compositionally and hence that concepts, which do combine compositionally, are not prototypes.

I owe this idea to Anthony Everett.

An analogous argument (focusing on the systematicity of thought) can be found in Gomila et al. (2012), who argue that the very systematic character of thought seems to depend on the systematic character of language. In particular, Gomila et al. argue that cognition is not by default systematic but rather thoughts are structured by the rules of syntax. Contra Fodor, it is argued that it is not the other way round. The combinatorial restrictions of thought are grounded in the combinatorial restrictions of the language used to express that thought. For instance, an English speaking person would not normally think ‘Cow brown’, while a Spanish speaker would. It is in this sense that systematic connections of thoughts depend on the syntactic structure of language and not on a Language of Thought (see also Tillas 2010 for a discussion). Furthermore, Gomila et al. argue that non-linguistic animals and subjects at early developmental stages (prior to four years of age) do not exhibit cognitive systematicity.

Further analogous problems potentially facing associationist views, like LASSO, concern homophones to the extent that they have to be disambiguated, pronouns whose referents have to be fixed, and so forth. The solution suggested for idiomatic usages also applies to these cases. See also footnote twenty-two.

The role of contextual features is perhaps more pronounced in cases of homophones. Consider thinking about your mortgage when hearing the word ‘Bank’. The chances are that you will not think about a riverbank but about the financial institute and an image of a (big) building or something along these lines will probably become activated. The same perceptual experience listening to ‘Bank’ would have been different if the contextual features were concerning an overflowing river. In cases like these, associations between different sets of words in each case play a key role for the meaning and in turn the usage of the word at hand.

Carruthers probably refers to arguments against verificationism and some sort of verificationist semantics.

The Sorites paradox concerns the indeterminacy surrounding limits of application of the predicates involved. For instance, starting from the claim that no one grain of wheat can be identified as making the difference between being a heap and not being a heap, it seems to follow that two do not either, thus three do not, and so forth. Eventually, it seems that no amount of wheat can make a heap, (Hyde 2011).

However, there is no direct evidence regarding the taxonomic abilities of patients whose naming loss results from semantic disturbances. But see Semenza (1999) for a suggestion that semantic disturbances would lead to impairments with taxonomic classification.

I am generally sympathetic to views of concepts as ultimately grounded in perceptual similarities, and I do not think that linguistic information is necessarily non-primitive as Roberson et al. (1999) argue. Further elaboration on this issue extends beyond the scope of this paper.

To the extent that one can talk about implicit judgements.

Their starting point is that the subtler the difference between two stimuli, the harder it is for subjects to classify them as visually identical or as belonging to the same visual category. Crucially, this was the case only when the second stimulus is presented slightly later, while the first is still present. The argument here is that this is an effect of the category label being activated during presentation of the first stimulus, rendering it more similar to the second stimulus that was presented a few milliseconds later. Response times were longer when subjects had to determine visual identity between two stimuli with subtle differences; these differences were eliminated between pairs of stimuli with subtle and less subtle differences when administering TMS disrupted the activity of Wernicke’s area.

See also Beale and Keil (1995) who argue that categorical perception is not limited to low-level processing but could occur at higher levels and could be acquired on the basis of experience. Baldo et al. (2005) argue for a role of language in complex problem solving via covert language processes. For further evidence for specific effects of language on perception, problem-solving, inferential processes and the effects of linguistic labels on memory, see Kay and Kempton (1984), Hunt and Agnoli (1991), Hardin and Banaji (1993), (reported in Sutton 2002).

Note that agrammatic aphasics could enjoy endogenous control over their concepts even in the absence of associations between words and representation of instances a given kind. For endogenous control could also derive through associating representations of instances to goal-directed states or other states over which subjects already have endogenous control, and so forth. See also the discussion in “Endogenously controlled thinking” section.

It is worth clarifying that by the age of nine to fifteen years both native and late signers can often ascribe beliefs to themselves and others in story narratives (Marschark et al. 2000), ToM reasoning difficulties of late signers can linger even up to thirteen to sixteen years of age (Russell et al. 1998). Both reported in Peterson and Siegal (1999).

I owe this suggestion to Gary Lupyan.

(1) Correct: phonologically accurate productions of the target. (2) Semantically related to the target (semantic paraphasias): any response that is semantically related to the target, such as “carrot” for “pumpkin”. (3) Semantically related description: description of the object with no attempt to name such as “you put coffee in it” for “thermos”. (4) Unrelated lexical error: responses that are unrelated to the target phonologically or semantically such as “key” for “pie”. (5) Phonologically related to the target (formal paraphasias): Word errors, which bear a phonological but not semantic resemblance to the target such as “pepper” for “pencil”. (6) Neologisms: non-lexical errors such as “lapita” for “blender”. (7) Perseverations: errors in which the word is used again from something said just previously. (8) No response (Farias et al. 2006).

Drawing involves spatial analysis and imagery, both of which are processes known to activate the right hemisphere, which is (largely) intact in most aphasic subjects. When an aphasic subject is confronted with a picture of an object, she observes it, a mental representation of the depicted object is formed in her short-term memory, and a matching process starts (e.g. Barsalou 1999). That is, a scanning process scans the subject’s long-term memory for a matching stored representation of previous encounters with an instance of the object in question. When a match is found, and a match is found given that subjects recognised the objects on confrontation, the stored perceptual representation, alongside the appropriate word, becomes activated in working memory. The activation of the appropriate word is what enables subjects, both aphasic and unimpaired, to name the perceived object. Once an unimpaired subject has named the object, and the object is hidden she can either recall the representation of the object and activate once again the word associated with it or merely recall the word (and along with it the associated perceptual representations of the object in question). Words cannot play the same facilitatory role in the case of aphasic subjects since they are simply unavailable to them. However, drawing facilitates naming performance in the aphasic subject by converting an endogenously controlled task into a stimulus-driven one. Given that short-term memory capacities of the aphasic subject are fairly unimpaired, [as shown from their performance under confrontation as well as from the work of Faglioni and Spinnler (1969)], aphasic subjects deploy their short-term memory capacities in order to recall an object. When the aphasic subject is asked to draw the object with which she has just been confronted, she focuses on the representation stored in short-term memory and tries essentially to copy it onto the paper in front of them. Judging from the quality of the obtained drawings, these representations, in the absence of constructional apraxia, are rather rudimentary. However, they still contain enough information from which the subject can benefit. That is, when the aphasic subject initiates drawing of the object that she has seen, some of the characteristic features/properties of the object are on the paper in front of them. This in turn plays the role of an external stimulus that replaces the hidden object. The process then followed is similar to the one that takes place in the subject’s mind under confrontation conditions. That is, the subject sees her drawing, a new representation is formed in her mind and a new matching process becomes initiated. When a match is found, the stored representation becomes activated along with the associated representation of a word. In this way, naming of the object is (to a certain extent facilitated), even though not systematically. Admittedly, the naming process is still compromised given the subject’s impairments, but this suggestion explains the performance of aphasics in the naming task under drawing conditions. Naming an object with which an agent is not confronted is an indicative case of recalling and thus a case of endogenously induced thinking. To this extent, the above experiments ultimately examined the ability of subjects to endogenously induce thoughts. In turn, the above results showing that drawing facilitates naming support the present suggestion that language does much more to thinking than merely communicating thoughts from the unconscious to the conscious level.

Impairment at the semantic level of language integration can be detected by asking patients to discriminate the meaning of a given word by choosing the object corresponding to the stimulus word from an array of semantically similar alternatives (Gainotti et al. 1983: 616).

Originally, Meili conducted this test by asking subjects to name the missing part. Meili’s target was to give instructions without using any verbal elements and hence to focus on the conceptual abilities of aphasics.

Based on the results of their experiments, Gainotti et al. (1983) suggest that further light could be shed onto Bay’s (1962) suggestions by stressing the correlation between conceptual and semantic-lexical disintegration. Semantic-lexical impairments in aphasic subjects are also significantly related to their inabilities to understand the meaning of symbolic gestures (evidence reviewed in Gainotti et al. 1983). Similarly, Gainotti et al. (1979) show that there is a strong relation between semantic-lexical disturbances and the inability of aphasics to appreciate relationships between pictured objects with different levels of conceptual similarity, e.g. chair and stool, bowl and cup, etc.