Before there was grammar
2.2 Where semantics begins
Where does‘semantics’ begin? Let us start with a broad sense of the term, where it covers the emergence of‘meaning’ long before words exist, and independently of whether there is communication. In any creature that has a brain at all, the analysis of sensory stimuli involves a transduction of such stimuli into neural signals, which transmit the information contained in the receptors to the brain. The brain then extracts information from the sensory signals and at least some brains will form ‘representations’ of environmental variables in some particular domain. These can in turn be structured, with different mental‘symbols’ representing different environ- mental variables, and formal computations can be defined over them. In the well- studied case of bee navigation, for example, the representations can represent such
variables as food sources, the time of day, or the compass direction of the sun as a function of the time of day (Gallistel,1998).
A structured formal semantics, mediating a perception-action circuit, is thus in place long before language arrives on the evolutionary scene. Ipso facto, a semantics of this kind will not tell us what is special about language or the organization of meaning in it—our basic question here. Nonetheless, there has been a temptation to think that when words came along, their semantics was already there: this, for example, would be so if human words ‘expressed’ prelinguistic ‘concepts’, which are their meanings. At this stage, therefore, we need to be more precise about what we actually mean by a word. First, there is the notion of the word as a prosodic unit (the ‘phonological word’); then there is the notion of the semantic word or lexeme, which is the word understood as an abstract vocabulary item with a given meaning that can take different forms, such as the verb run, which can take the forms runs, ran, run, etc. Even more abstract is the notion of a lexical root, which involves a semantic core possibly shared across lexemes of different categories, e.g. the root√run as involved in the verb run and the homophonous noun run, which occurs in the expressions a run, many runs, running, Mary runs, etc. Finally, there is the grammatical word: the word as a morphosyntactic unit or as functioning in a sentence context.
The second notion of ‘word’ appears to be virtually identical to what are called ‘concepts’ in the philosophical literature (e.g. Fodor, 1998), where indeed there is the same notational convention of referring to such entities with small capitals. Presum- ably, the words‘runs’ and ‘ran’ express the same ‘concept’—or at least, the question of whether they do seems barely, if ever, discussed. Since the technical notion of a lexeme is more precise, it is not clear why one should invoke the notoriously unclear notion of‘concept’ at all. That notion is often applied to non-linguistic beings (Carey, 2009), and if we do not wish to ascribe lexemes to the animals, then the term ‘concept’ has an advantage over ‘lexeme’. It appears, however, that we will lose important distinctions in this process. While meaningful categories in the sense above seem to be involved in both perception and in intentional reference, they are stimulus-controlled in perception, while being freely manipulable in intentional reference. When manipulated for purposes of reference, they are symbolic units that have a phonological identity. Hence lexicalization has taken place: the brain has created items that can be freely called up, partially independent of perception. This is unlike in the case of percepts, where there is no reference and no need for it, and no creative and systematic combinatorics.1
None of this takes the point away that perception in prelinguistic cognitive systems can involve classes of considerable abstraction, as when prelinguistic human infants individuate objects via sortal kinds (e.g. box, car, toy, cat) (Xu, 2005), a process that
1 The imagination is creative, but it is not combinatorial, it appears, in any systematic way, and it is not
is to some extent independent of the object’s perceptual properties. Objects that are all toys, for example, can be perceptually quite different. One of the most abstract of such sortals is object: whether something counts as an object in perception (or is identical to another) does not solely depend on its spatio-temporal properties. The same is true for sorts such as agent, cause, attention, or plurality, which again are, although not independent of perceptual features, only obscurely related to them and not apparently defined in such terms (Carey, 2009: chs. 5, 6).2If so, we might call them ‘concepts’, also because they are ‘inferentially rich’ in the sense that how something is sortally analysed enters into the child’s decisions on how to act (e.g. whether to reach for an object). However, although such representations differ from ‘perceptual representations in their abstractness and their conceptual content’, Carey (2009) stresses that this is in part a ‘matter of degree’ (p. 449). Indeed, like the abstract concepts of core cognition, the outputs of perceptual input analysers (whether an object is red, round, and distant, for example) are accessible to so-called ‘central’ systems as well, where they enter into inferential processes and determine actions such as reaching. Moreover,
like representations of depth, the representations of objects, agents, and number are the output of evolutionarily ancient, innate, modular input analyzers. Like the perceptual processes that compute depth, those that create representations of objects, agents, and number continue to function continuously throughout the life span. And like representations of depth, their format is most likely iconic. (Carey,2009: 450)
An iconic representation is an analogue one whose parts correspond to the parts of the object represented: the head of a tiger in a photo, say, to its actual head in reality. No word or sentence has meaning in this analogue fashion. Related to that, no issue of intensionality arises with iconic representations, as Fodor (1998) argues: an iconic representation of a giraffe cannot help being a representation of my favourite animal, if the giraffe is my favourite animal. When we use words, however, intensionality effects arise pervasively and systematically: thinking about animals identified as giraffes is not thinking about my favourite animals, even if they are my favourite animal. Representations in the linguistic format are also not reflexively and causally triggered by a perceptual input, as concepts in Carey’s sense are; and they are not modular, being capable of describing any domain whatsoever (Spelke,2003), even if perhaps imperfectly, as in the case of emotions; and neither percepts nor core concepts nor iconic representations are normatively constrained, in the sense that
2 As Carey (2009) stresses even for object kind sortals, although these are perceptually based, ‘it is
unknown what these bases are’ (p. 273). That I perceive you as a person, rather than as a surface or a physical object, does not appear to be explained by the visual properties of your surfaces, or your physical and spatiotemporal properties—even though, equally clearly, there are correlations between the physical properties of an object and the way we perceive it.
there are correctness conditions that they can fail to meet: indeed, we do not control perception. Neither percepts nor concepts based on them require reasons, and the question of truth does not arise for them. For illustration, consider the Heider– Simmel (1944) demonstration that normal subjects endow a movement pattern of animated geometrical shapes as depicted in Figure1 with rich social meaning, reporting percepts such as the large triangle ‘chasing’ the small triangle, shapes ‘looking for’ one another, ‘hiding’, ‘conferring’, being ‘furious’ or ‘frightened’.
Percepts of this nature, which arise without conscious control, systematically deteriorate or disappear when the movement pattern is slowed down or sped up. Subjects then report merely seeing shapes and geometrical paths again. This becomes the general situation with patients suffering from bilateral amygdala damage, who also describe the same animations in asocial and purely geometric terms, despite otherwise normal visual perception (Heberlein and Adolphs,2004). No questions of truth or correctness arise in either the patients or the normal controls.
Yet this perceptual semantics is rich, and explanatory questions arise that are completely unresolved even in this domain (never mind grammatical semantics). In line with this, Mausfeld (2011) stresses the ‘explanatory gap between the information available in the sensory input and the meaningful categories that characterize the output of the perceptual system’ (p. 159). As he illustrates, objects that we perceive are endowed with an abundance of material qualities. In the case of perceived surfaces, for example, these are captured in verbal descriptions such as‘soft’, ‘wet’, ‘malleable’, ‘silky’, ‘juicy’, ‘edible’, or ‘deformable’. Many such perceptual attributes correspond, in metaphysical terms, to dispositional and causal properties of the objects perceived. Again, the explanatory connection between such attributes and the physical or visual properties of the stimulus that trigger their perception remains largely obscure. We do not know where the semantic contents of our percepts and core concepts come from.
These problems deepen as we move up from perception to relational concepts such as cause, which David Hume famously noted have no basis in our sensory-motor ‘impressions’. There is similarly no perceptual basis for our notion of truth, or the relational notion of a‘predicate’. We perceive things as objects external to ourselves, but this construal is reflex-like and requires no concept of what an object is, nor
thoughts involving such a concept. Thinking in this sense is, unlike perception, not an‘online’ process: as we think, we don’t necessarily externally perceive anything or attend to what we perceive—most of which will typically be irrelevant to the contents of our thought. In turn, most things that are physically present and that we perceive, we don’t also name, when we see them—fortunately so, for there is so much that we see all the time, most of which is of no relevance at all to what we are thinking: for example, the font type of the words you are perceiving as you read this, or your mood, both of which you largely ignore as you try to come to grips with what you read. As you do, perceptual stimuli associated with reading, while processed, are thus switched to offline in conscious processing. The phenomenology of reading is crowded out by the phenomenology of thought.
Much of the latter phenomenology is accompanied by silent soliloquy, with sentences or fragments of themflicking through our minds incessantly, to various degrees of (usually silent) articulation—a barely controllable predicament that can annoy and keep us awake at night, requiring extraordinary efforts to be terminated altogether, as in long-term meditational practices. It is as if each of these thoughts cries out for our attention, not leaving us in peace, showing that linguistic thought of this nature tends to be inherently related to consciousness. That relation is very clear in actual speech itself, where we virtually always know that we speak, when we do, and indeed normally know what we say. Most normal utterances appear to be spontaneous rather than carefully planned and executed, yet the moment we make these utterances, we are consciously aware of them and their content (though we may get distracted by something we see or hear while we speak, including the quality of our voice). Both in soliloquy and actual speech, moreover, it is linguistically mean- ingful units that we are consciously aware of, such as words, phrases, or whole utterances, but not, say, arbitrary stretches of phonation within an utterance lasting exactly300 milliseconds, or three quarters of a word or phrase. The phenomenology in question thus obeys in part linguistic principles.
Not only does semantics systematically change as we move from perception to grammar, therefore, but conscious experience does as well. Consciousness presumably begins with perception—percepts are conscious virtually by definition—but then, perception need not be accompanied by percepts that we actually consciously process (like the letters you perceive as you read, when focusing on the content of what you read). Consciousness thus changes fundamentally as language intrudes and takes over our conscious attention, switching non-linguistic perception to‘offline’. This in turn changes our non-linguistic experience. If Helen Keller is any source, the sensory-motor experience of water is not the same when we can name it and when we cannot. With words, experience itself becomes a matter of reflection. Even if other species have a ‘mind’ or ‘think’ (in some senses of these words), only we may be aware that we do, which in turn influences what we think about: about thought itself, for example.
The source of this transformation in both semantics and consciousness plainly cannot be semantic: for semantics is what is being transformed. If‘semantics’ means ‘relations to the external world’ (perhaps specified causally), it cannot at present much illuminate or explain the contents even of percepts. The same applies to the phenomenon of intentional reference, in either acts of pointing or the use of words. A young infant pointing declaratively points, say, to an elephant, viewed as an objective entity that the adult also sees and understands in the same way, as falling under the same‘concept’. As words come along, the infant might mutter ‘elephant’ while pointing to the elephant, and in any act of reference that we make as adults, some meaningful descriptive category is involved in a similar way (for qualifications see below and Chapter 4). None of this follows from anything semantic. Thus, reference in non-linguistic animals takes a completely different form (and may well be explainable causally). Intentional reference, in the human sense, involves lan- guage, and it needs to be explained linguistically, not semantically.
When, under the causal control of an external trigger, monkeys or chickens call for alarm, the call can be acoustically distinct depending on what predator is perceived to cause the threat (Hauser,1996; Cheney and Seyfarth, 1990; 1998). But the great apes do not call for alarm as the monkeys do, which we expect if such reference is functional and apes simply are not under the relevant kind of threats. Alarm calls are not for conceptualization or reference to the world, but for ensuring that the right action is performed under the right external circumstances, expedited by a reflex-like, emotional reaction to a given trigger, with perhaps no underlying communicative intention at all. For that reason, the question of which‘concept’ is being expressed by any of these calls appears to be mute. Is it eagle, or into the bushes, or danger from the air? No concept appears needed because, even without one, the triangle between the caller, the stimulus, and the action is causally and adaptively closed. The calls occur when the stimulus does, which is also when the percept does, which identifies the predator. There is no evidence that the calls in question are ever applied to anything other than the stimulus, in the very adaptive context in question; or that the percept involved in one such call can be targeted as such, and be combined with another percept, a process that we argue requires lexicalization. Every one of the calls is complete in itself, requiring no further specification of what is to be done by the recipient (see also Bickerton,2009: 44–7, 68–9).3
Lexemes by contrast not only can be used referentially in the physical absence of their referent, but are also very incomplete in their meaning. The word‘eagle’ by itself does not denote anything in particular: not this eagle or that, not all eagles or some, not a kind
3 We second Bickerton (2009: 69) in his claim that the units of alarm call systems:
‘don’t really translate into human language. We can give an approximate meaning, or several possible meanings, in terms of our language, but the idea that underlying both is the exact same semantic expression is simply baseless.’
of bird as opposed to another, not the property of being an eagle, etc.—things that it can denote only once it appears in the right grammatical configurations. It is also used for purposes of reference and predication, in addition to being used as a directive for action, and it again requires a phrasal context, hence grammar, when it is so used. As a lexical item, moreover, it automatically stands in relations of hyponymy (an eagle is an animal, as is a leopard), and hyperonomy (there are several kinds of eagles). The existence of such a system of hierarchical relations is not driven merely by instinct, in the way that animal categorizations of environmental stimuli develop, as the facts suggest that it is subject to significant variation across speakers and languages.
The use of a lexeme is also never indexical in the sense in which animal alarm calls are. Not even words standardly classed as‘indexicals’, such as pronouns or demon- stratives, are indexical in the sense of the monkey calls. Thus the1st-person pronoun need not denote the speaker:‘I’ denotes the logophoric agent of the embedded clause in (1) below and the sentence subject in (2), hence in neither case the speaker. Moro (1997) makes a case that ‘io’ (‘I’) in (3) plays the role of the predicate (hence not of a referential expression). In (4), ‘he’ refers to whoever enters through this door, not a particular person. (5) can be said intelligibly by an academic speaking in Durham while currently employed as a guest professor in Barcelona:
(1) He said to me: ‘I love you.’ (Sigurðsson,2004)
(2) Kon K@gna n@-ññ yl-all Amharic
John hero be.PF-1sO 3M.say-AUX.3M
‘Johnisays that hei is a hero.’
(Lit.:‘Johnisays that Iiam a hero.’) (Schlenker,2003: 68)
(3) Sono io. Italian
‘It’s me’ (Moro,1997)
(4) He who enters through this door will be crowned. (5) Yes, I am there now.
Even so-called ‘indexical expressions’, then, exhibit intensionality and allow for displaced, predicative and descriptive readings, apart from their more paradigmatic referential ones. Even in the referential readings, a concept is always involved. Thus if I refer to myself as‘I’, I am plainly not conceptualizing myself in the same way as when I point to my body and say:
(6) I feel totally happy in this body now.
The referential behaviour of indexicals, moreover, as we shall see in more detail in Section4.4, co-varies with their grammatical behaviour. What we find again, then, is a correlation between grammar, lexemes, reference, and intensionality: these occur together. Correlatively, intensionality is absent and reference is most‘direct’, when no
content word is involved, as in interjections. In that case, emotional content is critically involved in the utterances, and there is very little grammar:
(7) a. Wow! b. Sh***! c. Hey!
Where there is no language or lexemes at all, any case for either ‘concepts’ or reference is very hard to make. As Davidson (2004) stresses, neither a ‘concept’ nor reference can sensibly be attributed simply because an animal makes an adaptively significant distinction. Sunflowers can direct themselves towards the sun without anything we would be tempted to call thought. To take a more complex case, cephalopod mollusks vary their appearance to suit the surroundings. To find the