Why Linguistics?
What no other species has
The science
Two opposite pulls
History
Novelty and vulnerability
Two models
Outstanding issues
In brief
References
What no other species has
Speech and language are rather obviously complex. The faculty and its universality make human beings special. There are other special talents which humans have. But the learning of speech and language starts earlier and proceeds to a common conclusion more completely and reliably than the learning of any other faculty. No other species communicates with anything like the compositionality of human language. The generality and exclusivity are rather obvious. But speech and language are developmentally vulnerable – the commonest sort of developmental issue. How is it that the learning process is both reliable and vulnerable? There seems to be a contradiction.
Imagine the claim on a set of instructions for some gadget: It always works, except when it doesn’t. Obviously absurd. How is the absurdity to be avoided?
The science
The science of speech and language is known as linguistics. Along with astronomy, linguistics is one of the two oldest sciences, beginning at least three thousand years in the middle East, India and Greece.
Everything we think or say to ourselves or to one or more others, we think or say in some context or situation, following some rules of assembly, with some sort of expression which may remain entirely within our minds, with some meaning.
Meaningfulness depends on the grammar. Traditional grammars listed forms, such as the differences between be, am, are, and is in “I want to be good”, “I am good”, ‘You are good” and “She is good”. This is traditionally regarded as the linguistics of English.
By a higher criterion, a grammar can define both what happens and what doesn’t happen, that neither “I is good” or “He am good” are English. By a yet higher criterion, we can seek to explain how the native speaker knows that in “He seems to want to be good”, the second -s in seems denotes a relation between he and ‘being good’ at the far end of the sentence. There is a non-trivial question about how this learning process mostly proceeds without effort with no necessary explanation in the first ten or so years of life. But it is only in the past 70 years or so that the question here has been asked. And formulating the grammar in such a way that this is a plausible process turns out to be no easy task. For any language, of which English is just one of seven thousand or so examples around the world, there are idiosyncrasies. But there are also universals. Unsurprisingly, the learning process is not always effortless.
Linguistics studies the various relations here, the speech situation under the heading of pragmatics, the rules under the heading of syntax, the physical expression under a set of headings broadly characterised as phonology, the meaning under the heading of semantics. Although all of the relations here are highly contentious, linguists mostly think that what they are doing falls under the heading of science, subject to standard principles of evidence and verification. Some, but not all, linguists think that what they are doing falls within biology, and call their linguistics biolinguistics. Others situate their work within psychology or sociology.
It is often noted that humans differ from chimpanzees by less than 2% of their DNA. But within that small percentage, one singular difference is human speech and language. With varying degrees of success, baby chimpanzees have been taught some signs from sign and artificial languages. But baby chimpanzee sign language is not like human baby sign language. It does not have the characteristic structure of human language, with both an infinite capacity and a commonality across all members of the species. On the infinity, sentences can be lengthened indefinitely, as by “What are you saying?” “I know what you’re saying”, “They think that I know what you’re saying”, “She suspects that they think that I know what you’re saying”, and so on. And there is commonality in the fact that this is reliably understandable. The understandability declines as the length and complexity increases. But there is no point at which the structure ceases to be English, as just one arbitrarily selected human language. No non-human has ever shown any awareness of the little word that in structures like “They think that I know what you’re saying”. In some languages, English being one, it is not pronounced in “They think I know”, but pronounced in “They think that I know ”. In English, that is optional. But the fact that it is specified by the grammar, whether pronounced or not, opens up both the infinity and the commonality. The device or devices which allow language to have these capacities is known as syntax.
By the age of three children are growing their lexicons by 10 or so new words every waking hour, more in a week than any non-human in a life time, and not just asking and understanding questions, but saying things like “I want to stand on the chair to see what’s happening”, with three bits of sentences embedded inside one another, with I understood as the subject of three of the verbs, but pronounced as the subject of just the outermost verb in the structure. I is is displaced twice, as the underlying subject ot see and of stand, One of the subjects of the research here said this just before he was three. The other subject said something similar at the same age. To explain this on the basis of a simple stringing together would be more complex than by repeating the same sorts of structure and then assembling them into one larger structure of the same fundamental sort.
The most profound clinically relevant lesson from linguistics is that there is such a thing as the learnability space. Some aspects of speech and language fall within it, and have to be learnt during every normal childhood. There are two interlinked aspects of what has to be learnt.
By one, the grammar has to specify all and only those phenomena that contribute to the faculty of language. The grammar has to do so in such a way that it is finitely learnable, and, by recent formulations, in such a way that it could plausibly have evolved. By this finite learnability, the learner learns that the order of adjectives, like good or bad, and what they modify varies according to whether it is a pronoun, like something, nothing, anything or everything which gets modified, or a noun, like song, idea, government. And this learning is robustly expected. A speaker saying “I see bad nothing” or “I like books good” would be barely understood, if at all.
By the second, the grammar has to be broken down into the smallest possible chunks. Such a grammar is both reductionist and ‘generative’ in the sense that it generates structures freely in two directions, by the speaker creating meaningful forms and by the listener assigning analyses to them. By Chomsky’s seminal 1957 proposal, Syntactic Structures, this breaking down was into two components, a ‘phrase structure’ component and a ‘transformational’ component. Within the latter, the rules manipulated variables. These could be added to, deleted, or reordered, giving what are recognisable as questions, negations, and so on, with all the elements minutely defined. These elements included the traditional category of tense, as in “She fell over” in contrast to “She falls over”. The novelty here was that all components were defined abstractly. The great empirical advantage of this abstraction is that it makes description precise. Some, like Ben Ambridge sneer at this on the grounds that it is ‘algebraic’. But without this precision, the finite learnability would not be possible.
Two opposite pulls
Especially from the perspective of speech and language therapy, linguistics is pulled or pressured in two opposite directions – in one direction to describe every language typologically and in the opposite direction to do so in the simplest possible terms, as by the biolinguistic approach, which I adopt myself.
The term, linguistics, only came into use during the 19th century.
But long before the terms linguistics came into use, the opposite pulls were evident – nowhere less so than in the study of the speech and language of children. This was the subtext of the bitter dispute about academic priority between William Holder and John Wallis in the 1700s. Their real disagreement would seem to have been about the order of definition in the study of speech sounds.
From my modern perspective, there are three criteria facing any theory of language: It should
- Describe any language precisely in the same terms which can be used to give an equally precise description of any other language, and do this in a way which is as clear about what is not part of the language as what is;
- Be learnable in the way language demonstrably is, so that, irrespective of the target language, the overwhelming majority of children develop an adult-like competence by the age of around ten without any specialised instruction; this competence includes, not just the broad outlines, but fine points exemplified in only a very small number of words, down to the limit case of one;
- Be such that the necessary capacity could have evolved in the human species.
The first two of these criteria were set out by Noam Chomsky in 1965. The third has become increasingly prominent in recent discussion.
History
Scientific thinking about speech and language is encapsulated in the history of the alphabet.. For instance, around two and half thousand years ago, there were various ways of showing the sounds of modern M, N, B and D, used by the Greek city states. Then someone seems to have had the idea of making the shapes of these four capital letters represent the sounds more consistently. Essentially, this was to represent the features of the sounds. Then people writing by hand in Latin started forming some letters in a style known as ‘minuscule’ or what we now know as ‘lowercase’, as opposed to ‘majuscule’, as used for public inscriptions. The same featural principle, as was used for B and D with the vertical left edge denoting a complete closure, was vaguely followed in the design of p, b, d, k, and with ascenders or descenders and opposite facing hooks and tails for q and g. These sounds, known as ‘stops’, all involve a complete blockage of the airstream in the mouth. Correspondingly more compact shapes were used for the five vowels in a, e, i, o, and u. The dot on the i and the cross on the t may have once represented a principle known as ‘minimal specification‘. By this (highly controversial) principle, i is the least specified vowel (in Southern British English and possibly in Latin) and t is the least specified consonant (in at least the overwhelming majority of languages). But if there were once featural design principles here they were not followed consistently. And then the featural idea was dropped until Alexander Melville Bell resurrected it with his ‘Visible Speech’ in 1867 with the phonetic features of speech which define the sounds reflected consistently in every character. But Visible Speech never caught on. And in othere respects the alphabet follows arbitrary, abstract, random, non-featural qualities.
Around the same time that the alphabet was being developed, a scholar named Panini, working somewhere in Northern India, wrote a formal description of Sanskrit in 3,959 succinctly stated rules, distinguishing consonants and vowels, nouns and verbs. Modern linguistics still embodies key aspects of Panini’s many insights. The most crucial of these insights was the notion of linguistic structures being ‘derived‘.
One of the first modern scholars of language was William Jones, who in 1786 set out the core idea of what are now known as Indo-European languages. Although the commonalities between Hindi and Western European languages had been noticed before, it was Jones who popularised the idea, defining criteria, still used today, by which commonalities could be identified, in the numeral system, in the system by which words are modified to form plurals like houses from house, known as morphology, and in the most familiar words. The scope of investigation was broadened to languages outside the Indo-European by Wilhelm von Humboldt who played a key part in stimulating American interest in the native languages of North America, not to mention his own research into Basque, and other languages now characterised as Austronesian, spoken across the Pacific. Gradually the coverage broadened. Eventually some six or seven thousand languages were identified across the world. The study of these is now known as ‘typology’. It is now apparent the Indo-European group of languages represents (by far) the largest single group in the world, having seemingly emerged around 6.000 years ago somewhere around the Black Sea.
The field changed fundamentally when Noam Chomsky became a student. His MA thesis in 1951 was a grammar of modern Hebrew involving thirty ordered rules. Crucially these rules were designed to describe not only what happened in the language, but also what did not happen. This degree of mathematical explicitness made Chomsky’s model different from any model which had ever been proposed before.
In 1957, Noam Chomsky proposed a grammar, partitioned into two components, one generating ‘kernel sentences’ like “The police interviewed the politician”, the other accounting for the very complex ways in which sentences could be ‘transformed’ as in “Who might not have been being interviewed by the police?” by a sequence of ordered transformational steps. Each component was characterised by a particular sort of rule. These defined questions (by words like who), tense (by -ed in talked), modality (by might), negation (by not), aspect (by have), the passive (by been), and more. Chomsky’s analysis was the first complete analysis of this core aspect of English, defining the interactions between these various aspects of grammar. Crucially, Chomsky showed that categories might be represented more abstractly than by words, as by talked, said and was.
In 1965, Chomsky proposed that a grammar of this sort was learnable by a ‘Language Acquisition Device’ which compared systems of rules, favouring the simplest. But in 1967 E Mark Gold showed that the class of languages assumed by Chomsky in 1965 was formally unlearnable. This forced a rethink about the defining architecture. The rethinking continues to this day.
In 1970 Chomsky made the form of word entries more significant and limited the scope of the transformational component. This reduced the problematic partitioning, and took the first step towards what became known as ‘X-bar theory’, generalising across sets of environments which had previously seemed entirely separate from one another, mainly nouns and verbs.
In the 1980s, particularly from work by Chomsky (1981) and Hagit Borer (1984), many linguists came to think of language learning in terms of choices between small sets of values with respect to particular variables, for instance whether or not the equivalent of I is routinely not pronounced, with the equivalent of “I love you” said without the I. This is the case in Greek, Italian and Spanish and numerous other unrelated languages. According to whether this happens or not, the language learner is thought to do the equivalent of throwing a mental switch one way or the other. Children exposed to English on the one side or Italian on the other would throw the relevant switch opposite ways. The points around which these choices or settings are made are known as ‘parameters’. From work by Nina Hyams (1986), this particular switch seems to be thrown at around two and a quarter.
Over a working life of more than 70 years, Chomsky has gradually refined his model, but without changing the concern for precision and explicitness. Increasingly he emphasises the notion which he sometimes calls ‘Computation for human language’ by which all of the seemingly very different languages of the world share one universal commonality. They all build their structures by the same GENERATIVE procedures – which DERIVE these structures in opposite directions for speech (or signing) and understanding. This model requires one universal semantics.
By current biolinguistic thinking, there is a key issue in evolvability.
Novelty and vulnerability
The great novelty of Chomsky’s approach has been to insist that grammars should generate only those structures corresponding to a set of canons making them grammatical, and not generate any structures not corresponding to these canons. By this approach, “Something good” is grammatical, “Good something” is not. The grammar should generate the former, and explicitly preclude the latter. In a more subtle way, in “He was seen to do something good”, the little word to has to precede do. In “I saw him do something good”, adding to before do is just not English. But “I want her to do something good” is fine. It does not seem plausible that this contrast is just stipulated in English grammar with respect to the word see (and hear which patterns the same way). A more plausible account is one based on general principles which separately or together have this effect. It remains an open question why these two verbs of perception should work this way – unlike want, know, force, and other verbs.
The contrast between grammaticality and ungrammaticality has been central to Chomsky’s work for the past 70 years.
Ultimately, the contrast here is by introspection, by writers (and readers) interrogating themselves. The slipperiness of the resulting data is obvious. But despite many attempts, it has proved hard to operationalise procedures to measure judgements across a sample of randomly selected experimental subjects about subtle contrasts such as the one between “I saw him do something good” and “I want her to do something good”. One of the many problems is to ensure that the task is understood consistently and uniformly. For all its obvious faults, introspection seems to be the only workable criterion. (See Frederick Newmeyer (1983) for a thoughtful survey of the issues here.)
Two models
Although Chomsky’s model is now followed by a very large number of linguists around the world, perhaps most, he has vociferous critics who argue that the generative model unreasonably seeks to force the observation of linguistic variety into a predefined framework. Many of the native languages of North America and Australia (of those that survive) have what seem to speakers of European-type languages like very elaborate procedures squeezing whole sentences into long words. Some of these are difficult to describe in generative terms. Julie Ann Legate did her PhD (2002) working on one particularly intractable case. There are claims, as by Daniel Everett, that they have found a language which cannot be described in generative terms. Such claims are always open to reanalyses of the same data. But if this proves impossible, the generative model is at least gravely weakened. Arguably it collapses. The stakes are high.
This difference came into the open in the 17th century, long before the notion of linguistics had emerged. The issue took the forms of a bitter, personal dispute (about academic priority of all things) between William Holder, who was, on a reasonable estimation, both the first generative linguist and the first speech and language pathologist, and John Wallis who took Holder’s place treating the same child, having previously given the first listing of the sounds of English. Wallis’s listing was strictly taxonomic, in contrast to Holder’s explicitly derivational approach. (Holder uses the term derivation in the modern sense). The real issue between them, which neither of them raises with any clarity, rebounds to this day. It is the seemingly simple matter of definition: What is a language? Is it useful to think of language as a series of categories, a taxonomy in other words? Or is this grossly insufficient? Should language rather be thought of in terms of the fact that the number of meaningful and grammatical sentences is demonstrably infinite? Generativists opt for the latter.
Outstanding issues
One of the issues between generative and taxonomic traditions is with respect to ‘correctness’. Generativists tend to pride themselves on the point that the starting point of linguistics is observation, not passing judgement. But taxonomists often respond that the very strength of the distinction between what the grammar generates and doesn’t generate is itself covcrtly prescriptive.
Another issue concerns the fact that some aspects of meaning are pronounced twice. And others go unpronounced. Consider the simple cased of who did what to who/ In languages like English, this is expressed twice, by the order of the words and by what is known as ‘case’. This duplication occurs in the simplest of sentences. In “She saw him”, it was she who saw and him who was seen. But these different roles are expressed not just by the forms of she as opposed to her and him as opposed to he, but also by the order of the words, the fact that she precedes saw and him follows.
By way of contrast, there is what is known as ‘elipsis’ where some of the structure is not pronounced. This occurs in the “I do” of the marriage vows. In some of the Celtic languages, such forms largely takes the place of English yes and no. But what can go unpronounced by elipsis varies across languages. So it has to be learnt.
There is another issue about how feisty, well-connected, working class women, completely unknown to one another, can all start saying the same speech sounds in some new way, and leading the speech and language community to follow them. This occurs in what is known as a ‘circular vowel shift’. Several such shifts are happening in North America right now (See William Labov (2001). Such shifts happen without anyone noticing because the steps involve fractional changes in the configuration of the tongue and lips for any given vowel. They are only measurable as ongoing phenomena with sound recording and computer data bases to process the great volumes of microscopic and subtle data.
There was a circular vowel shift in England roughly between the time of Chaucer and Shakespeare, giving the so-called silent E in bate, bite, note, Pete, from what were previously two-syllable words with the A, I, O, and E sounding quite different from the way they sound today.
A theoretical issue on which there have been significant changes of mind, is how to characterise contrasts in grammaticality or acceptability. It seems that there is a continuum between uninterpretable nonsense and what is understandable, but not quite right. The answer is not simple, certain, or obvious.
What is the relation between what is known as Universal Grammar or UG and discourse? This may have a bearing on the relation between what goes unpronounced and what is pronounced twice.
Linguistics is not a monolith. It does not pretend to answer all possible questions. It is a work in progress – particularly in the field of clinical linguistics.
In brief
Linguistics is cautious about what counts as fact. There is a tension between the three criteria above. They pull in different directions. They represent different priorities in research. But to my mind, the issue of evolvability is especially significant in relation to developmental disorders of all sorts, as outlined in Nunes (2002), subject to the corrections here.
With astronomy, experimentation is impossible, but it is possible to look into the future. With linguistics, experimentation is very difficult. There are limits to what can be learnt from psycholinguistics. But predicting what may happen next is just speculation.
In both linguistics and astronomy there is key data in both the recent and the ancient past. This bears directly on the three criteria above. Overcoming the tension between them is a key research objective.
Benefits
By the approach and assumptions here, speech and language therapy has everything to gain by taking account of the diversity, but not shying away from decisions where these have to be taken. Thus it cannot be the case that the notion of syntax is only half-right, that it properly fits within the broader framework of conversation. Either syntax exists, or it doesn’t. Only one of these ideas can be correct.
The benefits for speech and language therapy are profound.
References
Randolf Quirke (1972) drew attention to the therapeutic significance of this line of investigation as pioneered by Noam Chomsky (1957, 1965, 1968 and other work) and Eric Lenneberg (1969) with an appendix by Chomsky. But any inherited capacity is vulnerable. Hence speech and language defects and this website. In relation to the view defended here, see Adger (2003), Carnie (2021), Radford (2016), Roberts (2023)) for highly regarded introductions to current thinking in the framework assumed here, though not to the emerging issue of evolvability, which has only come to the fore since Berwick and Chomsky (2017), although the issue had already been under discussion since the 1990s.