Friday, January 2, 2009

My views on the NATURE-NURTURE debate – Part 4


Chomsky’s key arguments are as follows:
Speed of acquisition
Critical period
Poverty of input
Convergence amongst grammars
Language universals

Pinker bolsters Chomsky’s paradigm in many of his works, but The Language Instinct and Words and Rules will be considered in more detail than others since this is where the former outlines his case and shows us where he stand on this debate.

I will now explain each of these in turn:

Speed of acquisition

The fact that children acquire language so effortlessly and so quickly, despite its obvious complexity, shows that they must be innately disposed to do so. Other forms of learning are slow and protracted (cf. the freshman’s struggle in learning things like physics).

Stages of acquisition

Children learn language in the same stages all over the world, starting with the babbling stage, moving on to the one-word stage, then the two-word stage, then the telegraphic stage, and finally culminating in fluent speech.

Why would a child belonging to some tribe in the Amazon acquire his language in the same stages as those of an Italian child in Milan? In fact, children all over the world seem to follow the same pattern when learning a language. These stages occur in this order not because they are required to by the laws of logic, but because it is a matter of the unfolding of a genetic blueprint. If we do not accept this as the only viable explanation of the facts, we would either be forced to concede that we do not know, or have to postulate a rather surprising coincidence. If we accept that language is no more ‘learnt’ in this order any more than we “learn to have two arms”, it at least provides us with an intellectually satisfying explanation of the facts.

Critical period

Then there is the critical period thesis, which claims that there is a period in one’s life when one is particularly disposed to learning a language, after which language learning becomes so much more difficult. Chomsky insists that the “functioning of the language capacity is […] optimal at a certain ‘critical period’ of intellectual development”; just think of the mother-tongue English speaker trying to learn French at, say, university. According to Susan Curtiss, “there are two versions of Lenneberg’s critical period hypothesis”: the strong version and the weak version. The strong version says that it is not possible at all to acquire language naturally after puberty; the weak version says that one cannot acquire language with equal proficiency to that of a native speaker after puberty. As will be discussed in a subsequent section, the strong version is not testable in practice. Hence, the case of Genie is used to lend support to the viability of the weak version. (Nonetheless, I contend that both the strong and weak versions are flawed.)

Genie was the victim of an abusive father, who locked her up in a room upstairs. She was chained to a toilet seat, and not allowed to leave the room. They passed her food through a little flap at the bottom of the door. She literally had to live in a room for all her childhood. The neighbours reported something strange in the household, and after investigation, a social worker eventually discovered Genie.

Upon her discovery, the father shot himself dead, and her mother claimed that she was continually beaten into submission to follow his orders, and therefore could not do anything about her quandary.

Without going into the maudlin details, this story is relevant because when they discovered Genie she could do little more than feed herself. She walked in a very animal-like fashion, had no sense of decorum, often grunted as if it was normal, and she could not speak or understand language. After intensive therapy, she was able to control her outbursts, stopped grunting, learnt how to walk properly…etc. However, up till today, she has never learnt how to speak with the efficiency of a native speaker. Despite some of the world’s best psycholinguists and speech therapists working on her, she was only able to utter very basic, rudimentary sentences. The question is: why did she (inter alia) learn how to walk properly, but could not acquire language proficiently? The nativist’s answer: because she had passed her critical period for language acquisition.

Poverty of input

Chomsky claims that the input children get from their linguistic environment is far from ideal, and children are generally not corrected on grammatical issues - parents are more interested in things like ethics and truth-content. Even though overt correction is not necessary, it is still interesting that even young children never utter sentences which are syntactically deviant, even though they may over-generalise rules. Chomsky would say that this is because we have the syntax hard-wired into our minds.

The point is that children must infer general rules about language after hearing a few existential examples, despite not being told to do so, and despite the fact that even these examples are not ideal.

Aside from the anecdotal evidence that Chomsky presents, he always points out children’s knowledge of yes-no question formation in complex sentences. He explains that if we consider a sentence like

The boy who is at the shop is naughty

there are two ‘hypotheses’ a child could construct when trying to work out the rule, let us call these hypothesis 1 and hypothesis 2:

Move the first verb of the sentence to the beginning when forming a yes-no question
Move the verb that occurs in the matrix sentence (as opposed to the embedded clause) to the beginning

Of course, the correct rule would be hypothesis number 2, which would produce the grammatical sentence:

Is the boy who is at the shop naughty?

However, the point Chomsky is trying to make is that no child would ever utter a question using hypothesis number 1:

*Is the boy who at the shop is naughty?

If the child was going on what he heard from people around him speaking, there is nothing in the speech input that prevents him from distinguishing between the two hypotheses. So, given that the application of hypothesis number 1 is a logical possibility, and given that number 2 is also a logical possibility, what precludes the child from forming questions at least some of the time in the manner above, using hypothesis 1?
If the child is using the “fragmentary input” from the environment as a basis for his hypothesis formation, there is absolutely nothing that stops him from doing so. Logically, then, Chomsky would have us conclude that the child is not using the environment, but comes with his own innate linguistic structure hard-wired into his mind. This might seem prima facie implausible for someone using his common-sense, but how would you otherwise explain the above-mentioned facts?

If one considers the structure of the affirmative sentence, it would look like this:







The boy who is at the shop is naughty

As this structure clearly indicates, there is a relative clause which is embedded in the matrix sentence. Now, when forming a question, the verb that needs to move to the beginning is the verb that occurs as part of the VP that is higher up in the tree structure, i.e., part of the main VP. For that reason, the only way to form a question from this statement would be to move the main verb to the beginning, not the one that happens to appear first in the linear order of words, because it is much lower down on the structure due to its being part of the subordinate clause.

Because the transformation can only occur in this way, it would result in a structure like:







Is The boy who is at the shop (trace) naughty

Without knowledge of the structure, there is no way a child could know this. Given that children are not taught about the structure of language, we would have to explain how it is that children seem to always make use of principles like this.

If we do not concede that we have this elaborate linguistic structure hard-wired from birth, then we would not be able to explain why many aspects of even very young children’s linguistic competence is “grossly underdetermined by the fragmentary input from the environment”.

Convergence amongst grammars

People in a community all speak their mother tongue with equal proficiency, regardless of educational background; a ten-year-old child speaks with the same grammatical complexity as an educated adult. Given all the logically possible ways a language could develop, it is rather surprising that linguistic competence would not be as disparate as one would expect if there were no innate mechanisms constraining the development of grammar.

Language universals

In this section I will be discussing universals insofar as they relate to structure-dependency in language. I have not included the other kinds of universals that Chomsky refers to because the evidence for them is so weak that they are not even worth considering seriously. To illustrate what I mean by this, I will provide just two examples:

Colour Terms - The claim that all languages segment the colour spectrum, such that we have different words referring to different colours, is something that needs to be explained, since that is not the way things are in nature; the various colours simply ‘blend’.

Hence, how does one account for the fact that some languages have “one word for both blue and yellow”?

Related to Chomsky’s point on colour terms, Berlin and Kay conducted a survey, published in 1969, of ninety-eight different languages, and noticed certain patterns regarding their use of basic colour terms. By basic, they mean terms that are morphologically simplex (which precludes terms like bluish), non-compound terms, and terms that are not ‘shades’ of another colour ( thereby precluding terms like scarlet, which can be considered a shade of red), and words with restricted applications (like blond, which only refers to a male’s hair-colour).

According to Berlin and Kay, languages differ in the number of terms they utilise, but not otherwise. They made the startling claim that languages follow a universal schema. This schema can be illustrated as follows:

black pink
red green blue brown purple
white yellow orange

What Berlin and Kay claimed is that if a language has two colour terms, they would be black and white. If they had three terms, they would be black, white and red. If a language had four terms, they would be black, white, red, and either green or yellow, etc. In languages which have more than seven terms, the schema breaks down, but the interesting thing is that in languages with seven or less terms, the schema applies universally.

According to Geoffrey Sampson, these results are only impressive if taken at face-value. Firstly, the results were not gathered by Berlin and Kay themselves, but by their students at the University of California. Each student was to find a language, and find out as much as they could about its colour terminology as part of their course work. Perhaps it is for this reason that some of their findings were inaccurate. For example, there are four basic colour terms in Homeric Greek, one of which is glaukos, which is actually translated as ‘gleaming’, with no implications of colour. Later on it came to mean something like what the English derivate ‘glaucous’ means: bluish-greenish-grey. However, they needed a word for black and duly translated it as such, despite the fact that there was a standard word for black in ancient Greek, viz. melas, and in the Homeric texts, it is the most common colour term.

With regards to loan words, Berlin and Kay include them when it fits their schema, and exclude them when it does not. What ought to happen is that they should decide at the outset whether or not it is viable to accept loan words as part of their data, and stick to that. However, they exclude the Chinese loans from Korean, but include the same Chinese loans in Vietnamese. This is because the latter are more amenable to their theory; the former are not.

Nevertheless, these are just some of the problems with the treatment of Berlin and Kay’s theory of basic colour terms. Sampson’s critique is more comprehensive, but the point I am trying to make is that this theory is not at all as water-tight as commonly assumed.

Proper nouns –Chomsky claims that there is “no logical necessity” for names to meet any condition of spatio-temporal contiguity, but the fact that they do is most certainly “non-trivial”, and in need of explanation. For example, the fact that no language makes use of a word like LIMB to refer to a quadruped’s four legs as if they were a single object (as in “The cat’s LIMB is grey”, referring to its four legs) needs explanation, because there is no logical reason for this restraint.
Regarding spatio-temporal contiguity, consider a noun phrase like The Three Sisters, with reference to the constellation of three linear stars, which are actually millions of kilometres apart. Also, French has a word, rouage, which means a singular object comprising the wheels of a vehicle. These are counter-examples which clearly show that these restrictions are actually not restrictions at all.

Examples of this kind are numerous, but due to the fact that just about every alleged universal has been refuted, or as Jackendoff once euphemised it, is “in need of explanation”; hence, these concrete examples will not be dealt with any further. What are more interesting are the abstract properties that languages have in common, or more specifically the fact that languages make use of hierarchical organisation, i.e., structure-dependency. This is interesting because:
- It is quite possible, logically, for languages to be structure-independent, and one can easily imagine a possible world where this is indeed the case.
- It is a matter of fact that all human languages so far known to us use structure-dependency, and this needs to be explained.

Chomsky says that if a Martian scientist were to come to earth to study our language, he would conclude that Earthlings speak a single language. The point he is trying to make (Chomsky, not the Martian scientist) is that all the languages in the world have a similar underlying structure, despite superficial disparities. The languages of the world are not as diverse as they could be because they share properties which they are not logically required to share. The fact that these findings are empirical claims is what makes this strand of Chomsky’s argument so convincing.

Many contend that the argument from universals is one of the strongest supporting the nativist’s paradigm. In fact, Chomsky has often remarked that the goal of modern day linguistics is to find out what these principles are; sometimes this is referred to as “the initial state of the Language Acquisition Device” (which is what Universal Grammar is) by Chomsky, or more simply the “language instinct”. Chomsky’s argument is by and large syntactically motivated.

It may be worth noting that in his earlier works, Sampson was quite content to agree with Chomsky’s inference of innateness from the fact that languages make use of tree-structuring. In a book called LIBERTY AND LANGUAGE, Sampson he says that “until someone offers a concrete and demonstrably preferable alternative, we will do well to believe the standard explanation… [and] I believe that Chomsky has quite adequately made his case for genetic inheritance…” (p. 24 – my italics). Of course, Sampson has now rejected even that, but we will deal with Sampson’s critique in a subsequent section.

Chomsky’s universals are syntactically motivated. The syntactic universals that Chomsky has pointed out have to do with the hierarchical organisation of sentences. A sentence consists of a main clause within which will typically be included subordinate clauses of various types. A clause has a verb at its core. These clauses are made up of phrases, which have a head and optional modifiers. Hence, an NP has a noun as its head, which may be modified by an article and adjectives. In other words, any sentence includes a hierarchy of elements of different sizes, groups of which are nested within elements at the next higher level of the hierarchy, and each of which is decomposable into a group of elements at the next lower level.

Chomsky illustrates how this hierarchical property of language is not logical by contrasting it with fictitious languages that are structure-independent. So, let us consider question-transformations, for example. A question is formed from a statement by moving a certain word to the beginning, so The men John wanted to see are in the kitchen becomes Are the men John wanted to see in the kitchen? The word to be moved (are) is chosen by reference to its position in the hierarchical structure: it is always the main verb of the main clause of the sentence. (This is discussed in more detail below when I talk about Sampson’s response to the ‘poverty of input’ argument). The point is that it would be perfectly possible to imagine languages whose questions were formed according to some principle that does not incorporate structure-dependency. Such a language might have, for example, as a rule, something like “To turn a statement into a question, front the third word in the statement.” No language does this, even though it is a logical possibility. There are countless other hypothetical rules that can be constructed, like “to form a question, reverse the order of words”, but we will not state them all explicitly. The point is that all the (known) languages of the world make use of hierarchical inclusion relationships, and that needs to be explained. This is a contingent fact that does not follow as a logical necessity.

Ironically, Sampson himself, in his earlier days, once conceded that this discovery is tantamount to discovering something like: the number of thorns on a rose bush is always a multiple of seven. This is an empirical finding, and needs to be explained. Why would one not agree with someone who says: the fact that every rose bush all over the world, regardless of the environment in which it grew, always displays this property? Does it not surpass belief to think that it could be due to mere coincidence? Hence, he accepted Chomsky’s idea of innate linguistic knowledge on that basis.

Upon analysis, we do indeed find that the various languages display certain universal features which follow as a consequence of structure-dependency. In addition to that of the question-transformation rule, let us consider another example, given by Chomsky:

a) John said that he was sick
b) He said that John was sick

In a), there is an ambiguity; it is not clear whether ‘he’ refers to John, or somebody else. In b), by merely inter-changing ‘John’ with the pronoun ‘he’, we know that ‘he’ cannot refer to John, but has to refer to some outside entity instead. One gets a better idea of why this happens if we consider the structure of these sentences. a) would look like this:

[John] [said that he was sick]

whereas b) would look like this:

[John [said [that [he [was sick]]]]]

The reason is that in a) ‘he’ can refer to the NP John because ‘John’ is “higher up” in the phrase structure tree than ‘he’. In b), ‘He’ is part of the main NP, making ‘John’ part of the NP that is lower down (hierarchically) on the phrase structure tree; and anaphors cannot be bound to a referent that is lower down in the tree. (The whole point of phrase-structure trees, by the way, is to show hierarchical inclusion relationships. Given that even four-year-old children know these things, knowledge of this kind must be part of our genetic make-up, since children are clearly not taught things like this. It should come as no surprise then that every native speaker of English intuitively knows that b) is dis-ambiguated (though most would not know why), given that we have these phrase-structure rules pre-programmed into our minds. If we believe that language is only acquired from the environment in which the child is in, “it seems inevitable that major aspects of verbal behavior will remain a mystery”.

(Unfortunately I cannot illustrate the tree here, as this would have made the point more clear.)

Chomsky thinks there are many other principles which constrain the structure of language, and that the goal of contemporary linguistics is to discover exactly what the nature of these constraints are. Some tentative principles which he postulates are the Extended Projection Principle (EPP), and the principle of full interpretation. The above-mentioned example is an instance of the binding principle.

Chomsky says that it is not the case that structure-independent principles are too complex; in fact they are cognitively quite simple. “Rather, what seems to be going on is that the messages humans need to transmit to one another through the medium of language are messages that come in certain fixed forms, and these forms to a large extent dictate the syntactic patterning of human languages”. These invariant syntactic properties are determined by the mechanisms of genetic inheritance.

The obvious, common-sensical objection to the thesis that we have a Universal Grammar is: how do we explain the fact that not all languages are the same? Chomsky accommodates this disparity into his theory by explaining it in terms of “Principles and Parameters” (henceforth P&P).

This theory claims that the child is able to learn language so effortlessly because he is born with a set of principles, together with a “list” of (unset) language-specific parameters. When these parameters are set, these, together with the common principles that form part of our biological endowment, constitute what Chomsky calls “knowledge of language”. So we are born with rules governing sentence structure, like (to oversimplify) the following phrase-structure rule:

S -> {NP,VP}

The comma indicates variable order. When the child is about four years old, he realises that the language he is exposed to, language X, has the NP-first construction, and the genetically pre-programmed Language Acquisition Device crystallises this rule to read:

S -> NP VP

In language Y, however, the child realises that his language employs the VP-first construction, and now the genetically pre-programmed Language Acquisition Device crystallises this rule to read:

S -> VP NP

Hence, these two children may grow up speaking mutually unintelligible languages.

Obviously these are not the only rules that get crystallised during this process of so-called “parameter setting”, but we will leave it at that for the sake of simplicity.

This is the nativist’s explanation as to why language can be both innate, and variant across the human race. We will see below that this theory has problems which need to be addressed.

Nativists are wont to ask why we cannot form questions by interchanging the first and second words in a sentence, by reversing the order of words, or by picking out the middle word in a sentence.

What is wrong with turning the sentence

John ate an apple

into a question by saying

Apple an ate John?

Why does this feel so very unnatural? As pointed out, this is because language always makes use of structure-dependency. But if it is possible in principle for a language to have a rule like “Reverse the words in a sentence to form a question”, the fact that it never occurs needs explaining. This mystery can be explained if we assume that this restriction is hard-wired into our genes, and therefore restricted by our biology. If we do not accept this, then we cannot explain why the above-mentioned rule is so unnatural, and we would have to be at a loss in terms of being able to explain why rules of this type are so uncomfortable for us to utilise.

Descartes’ Problem

One can outline what Chomsky refers to as “Descartes’ problem” (cf. his book CARTESIAN LINGUISTICS) as follows:
How is it that we as native speakers are able to understand and produce a potentially infinite number of sentences, despite very often hearing/producing utterances that one has never before encountered in his life, or as Chomsky puts it, never before in the history of the universe. This Chomsky refers to as the creative aspect of language use (aka “Descartes’ problem” – so-called because it was Descartes, according to Chomsky, who alluded this problem).

In addition to this, Chomsky emphasises the fact that speakers are not able to express this variety of linguistic utterances due to being endowed with the relevant physiological apparatus (vocal tract, etc.), because a parrot is able to do the same, and he contends that the physiological mechanisms pertaining to the manifestation of sound are not necessarily in any way inferior to those of a human beings.

What, then, distinguishes a parrot’s ‘speech’ from human speech? Chomsky tells us that it is because there is something unique to the human mind which neither parrots nor any other (non-human) animal possesses. This ‘something’ is none other than our inborn Language Acquisition Device.

Aside from the fact that our innate rules give us the machinery to be ‘creative’, humans also have the ability to tell which sentences are grammatical, and which are not. How are children, faced with countless possibilities as to how to string sentences together, able to actually make sense of what they hear? If children were going on what they hear, they would need some way of knowing which sentences are acceptable, and which are not, and given that the possibilities are endless, how would children be able to cope with this mammoth task of learning the rules of language? Once again, Chomsky says that this would not be a mystery if we simply had a genetic predisposition to do so.

That pretty much sums up Chomsky’s case for his theory. His other writings serve only to supplement the points made above, with minor variations.

Let us now turn to the work of Steven Pinker, one of the leading exponents of neo-Chomskyan nativism. Pinker’s point of view will be taken primarily from his book THE LANGUAGE INSTINCT, and less so from a later book called WORDS AND RULES.

The Language Instinct

Here now are some of the key arguments found in Pinker, followed by my criticisms:
Phonotactic Constraints

Pinker claims that languages have rigid rules about how sounds are assembled into words. Many introductory texts call these “phonotactic constraints”. If you had to ask a native speaker of English about the grammaticality of certain words, he would somehow always know whether they are acceptable words or not. For example, words like “plaft” and “blick” are unequivocally judged as possible yet contingently non-existent words. Words like “sram” are unequivocally judged as impossible, and necessarily non-existent. This is because “plaft” and “blick” fit in with the existing sound system of English, whereas “sram” does not; it has an illicit onset sequence of “sr”. These constraints are part of our LAD, which allow us to set the parameters for our language at some relatively early stage of our linguistic development.
Infant Prodigies

Pinker says that children have to know what to expect when acquiring language, otherwise they would take forever to learn their mother-tongue. How would they know what to test for? How would they know what a verb’s inflexion depends on? Quoting Pinker, Sampson says that logically speaking, “an inflection could depend on whether the third word in the sentence referred to a reddish or a bluish object; whether the sentence was being uttered indoors or outdoors, and billions of other fruitless possibilities that a grammatically unfettered child would have to test for”.

Pinker then goes on to say that children never make the mistakes we would expect them to make if they learnt language from scratch. Pinker also refers to one of Stromswold’s Stromswold’s studies, based on the speech of pre-schoolers. She found out that even though there were false but logical analogies that could have been made by the children, they were never tempted to give in to these mistakes. Why does the child not commit a false analogy like He cans go based on I like going → He likes going; I can go → ???

Stromswold reported that the pre-schoolers never made a single false analogy. It is not the case that children never make mistakes when acquiring language. Pinker himself quotes the example of a father trying to correct his daughter from saying other one spoon to saying the other spoon, but to no avail. So what does Pinker want us to conclude from Stromswold’s work? It seems as if he is saying that the rule placeholders, such as ‘one’, can be followed by the noun for which it stands is a rule found in some languages, and since the child has not quite set her parameters to English yet, she utters possible, but (contingently) non-occurring utterances. However, no child will try out the idea that “can” will inflect to agree with its subject if it is in the third person; our innate knowledge precludes this kind of option, therefore children never commit this error.

Second Language Learning and the Critical Period

Pinker mentions Johnson and Newport’s research on Korean and Chinese immigrants to the USA. They found that the later the immigrants arrived, the less proficient they were at detecting grammatical errors. They also found that the poor immigrants’ attitude towards the language, and their motivation, had nothing to do with it; their age when they arrived had everything to do with it.

Pinker also mentions that despite the fact that Meryl Streep is renowned in the US for her accents, her English accent is “considered rather awful” in England, and that her Australian accent does not “go down too well in Australia either”. This is the case because she is trying to speak like a native speaker when the rules for her L1 have already crystallised, and she will therefore never be able to speak as proficiently as a native speaker.

The Language Mutants from Essex

One seemingly compelling case is that of one family who exhibited an anomaly known as SLI (Specific Language Impairment), whereby certain linguistic faculties are impaired, and their cognitive functioning in other areas were normal; the fact that it was carried over within the family shows that there certainly is a genetic element involved. Despite the fact that many of the ones who had SLI grew up in a “linguistically normal” environment, they still exhibited the anomalies that SLI sufferers have. Pinker also reminds us that they are cognitively normal in other domains.

The Structure of Words

Pinker claims that the structure of words also lends evidence to nativism in the sense that children’s minds seem to be designed with the logic of word-structure built in.

His first example is that of a pattern whereby the derivational affix can attach only to roots, not to stems (i.e. roots with at least one affix attached). When we try to attach a derivational affix to a stem, we get something like Darwinsian, which could refer to the works of Charles Darwin and his grandfather, Erasmus Darwin, whose work Charles drew profound inspiration from. However, this would sound ridiculous, not because it is illogical, but because there seems to be some sort of innate restriction on what affix can go where in a word. Later Pinker says that the restriction is that derivational suffixes cannot attach themselves to inflected roots.

Pinker also discusses headless compound words, and draws conclusions which provide further evidence for his case. An example of a headed compound would be barman, with the –man part constituting the head. Hence, a “barman” is a kind of man, viz. one that works in bars. Headless compounds are words like “walkman”. Here, -man cannot be the head because a “walkman” is not a kind of man. (Nor is it a kind of walk, but the rightmost morpheme is usually defined as the head of a compound).

Pinker’s explanation is that when a compound has a head, all the information regarding the head gets percolated up to the top-most node of the (morphological) tree, including the rules. When there is a headless compound, it is not clear where the information must come from, so the “percolation conduits” get blocked, and the default rule applies. That is the reason why we say “sabre-tooths” and “still-lifes”.

Other research cited by Pinker includes Peter Gordon’s experiment with 3-year-olds, commonly referred to as the “Jabba experiment” (Gordon, 1986). He showed them a puppet and said: this is Jabba. Jabba eats mice. Jabba is a ___? And the children reply: Mice-eater. Thereafter, he asks: Jabba also eats rats. Jabba is also a ___? And the children reply: Rat-eater.

Now, Gordon asks, why do they use the plural “mice”, but not the plural “rats”? Given that not a single child ever used the inflected plural as the initial constituent of the compound, this could indicate that there is some sort of rule that they are applying. It is unlikely that they were told that this is so, as Motherese contains few or no compounds, especially of that kind; and it would be preposterous to assume that someone told them not to include inflected words in their compounds. Why do they not generalise the rule: “create a compound by putting two words together”? Why should there be a restriction of this sort? The nativist would explain it as follows: if there were an innate rule of this kind to facilitate language acquisition, the child would already “know” some of the things that cannot be said, like including an inflected word in the first part of a compound word, which is obviously why they avoid constructions of this sort.

The results were interesting because there seemed to be a consistent restriction in terms of what kind of word was allowed to be in the first part of a compound word. “Mice-eater”, for example, was an acceptable form to them; their answer to the first question would be just that. However, they never answered “rats-eater” to the second question; they always answered “rat-eater”. This is interesting because if one goes solely on logic, then “rats-eater” ought not to be precluded, since “mice” and “rats” are both plural forms, and both (equally) morphologically complex.

The question, then, that Gordon posed is: what exactly makes forms like “rats-eater” ungrammatical? The reason he postulated was that the form “rats” is clearly derived from a rule: rat + -s. The word “mice”, however, is not; despite its morphological complexity, the word is learnt as a listeme, and therefore treated as such by the lexicon. Hence, “mice” in “mice-eater” is treated by the phonology as a simplex word.

This, Gordon claims, is an innate constraint placed on us by UG, and we have fairly good reason to assume this, given that children as young as three consistently do it.

There are problems with this claim, however, since there are languages which do exactly that which is claimed by Gordon to be universally excluded by an innate constraint. German, for example, makes use of regular plural forms in compounds. Consider the following set of data in the tables below:

English compound
German translation
1) Mice-eater
2) Rat-eater
3) Butterfly-eater
4) Insect-eater
5) Cat-eater
6) Flower-eater
7) People-eater
8) Tree-eater
9) Sheep-eater
10) Fish-eater

In German, -e(n) represents the plural suffix. To clarify, let us look at the singular/plural forms of the (German) words used in the data above:

1) Maus
2) Ratte
3) Schmetterling
4) Insekt
5) Katte
6) Blume
7) Mensch
8) Baum
9) Schaf
10) Fisch

In examples 9) and 10) [schaf(s)fresser and fischfresser], the singular form was preferred by my German informant; in examples 1) and 4) [mäusefresser and insektenfresser], both the singular and plural were judged to be acceptable forms, such that both mausfresser and insektfresser would not count as ungrammatical, though mäusefresser and insektenfresser are the more preferred of the two.

What is interesting about this is that examples 4-8 make use of the plural form in exactly the way Gordon would predict to be impossible, since here we have rule-derived words found in the position they should not be found in.

These alleged counter-examples can be dealt with if it can be shown that the phonology treats these morphologically complex words as simplex; in other words, the phonology is blind to the morphological complexity of these forms. Now we know that irregular plurals are certainly treated by the phonology in the same way as simplex listemes, but if we can show that this is true of not only irregulars, but other complex forms as well, maybe we can save Gordon’s thesis from these alleged counter-examples.

Jonathan Kaye, a phonologist from SOAS, has given us some novel insight into this issue of word structure. In discussing Kaye’s views, we will see why words cannot always be classified as either simplex or complex a binary fashion. Hence, let us now turn to Kaye's discussion of this issue (his theory is based on Government Phonology), so as to gain a better understanding of the matter at hand. (Note: the data here is once again based on English):

We cannot assume that the phonology is per se blind to morphological complexity, since the [mz] (from "dreams") is clearly a pseudo-cluster, and deemed both morphologically and phonologically complex.

It is clear, then, that at least some morphological structure is visible to the phonology. Kaye draws a distinction between what he calls analytic morphology and non-analytic morphology. Analytic morphology is morphology wherein the phonology is able to recognise a complex domain structure; non-analytic morphology is not able to. Let us first consider analytic morphology (which is visible to the phonology).

Analytic Morphology

As mentioned, the [mz] cluster from “dreams” is invariably complex. What does this tell us about morphological penetration into phonology?

Let us firstly assume that only morphological domains can be represented in the phonology. A compound like “black-board”, then, will have the following structure:

a) [[black] [board]]

Since there are three pairs of brackets, it reflects three phonological domains. These brackets represent how the phonological string is to be processed. To give an exact definition, we can use the CONCAT and φ (apply phonology) functions, so “blackboard” can be shown as

b) φ (CONCAT (φ(black) φ(board)))

To translate this pseudo-formula into plain language: apply phonology to “black”, then to “board”, concatenate the result, and then apply phonology to that string. So the brackets we see in example a) are actually not part of the phonological representation, but actually outline domains which are arguments to functions.

So this illustrates one type of structure that involves morphologically complex forms. It has the schema [[A][B]]. This is the case for most English compounds.

There is a second structure, involving two morphemes, but only two domains. The structure is [[A] B], and is interpreted as follows:

φ (CONCAT (φ(A), B))

To translate, we do phonology on A, and concatenate the result with B; then do phonology on the result. This is the case with English (regular) inflexional morphology. For example, consider the past tense of the word “seep”. Its structure is [[seep] ed]. With regards to Government Phonology, we could also illustrate this structure in more detail:

[[O N O N] O N]

x x x x x x x

s i: p d

Note that empty nuclei are to be found at the end of each domain, and are licensed to be silent by virtue of their domain-final position. Also, the licensing principle states that all positions in a phonological domain need to be licensed, unless that position is the head of that particular expression. This is actually why the suffix does not form a domain by itself. We see here that the suffix consists of two positions: an onset position and a nuclear position, both being licensed. If “-ed” were a domain, it is obvious that it would violate the licensing principle, which states that a phonological domain must have at least one unlicensed position – its head. The onset is licensed because, like all onsets, it is licensed by the nucleus to its right, which is itself p-licensed, i.e., licensed to be silent .

Analytic morphology also has a rather interesting property in that the distribution of empty nuclei is very restricted; in English, it is almost always found at the end of domains. Hence, this provides us with a fairly reliable parsing cue, since a p-licensed empty nucleus signals the end of a domain, so if it occurs within a word it shows that it is not a morphologically simplex one. So the pseudo-cluster [mz], from “dreams” is so-called because:

The vowel length is maintained (which would not be possible had it been an authentic cluster), and [m] and [z] are not homorganic, which again is not possible for true clusters. It is for this reason that much of analytic morphology is phonologically parsable.

So hitherto, we have considered two structures:

[[A] B],

and we have seen how morphology impacts on phonology in the form of these two domains.

Let us now turn to non-analytic morphology (which is invisible to the phonology).

Non-analytic Morphology

If morphological structure was present in the form of domains, then to that extent morphology can have an impact on phonology. These domains have the effect of respecting the integrity of the internal domains. By this, Kaye means that when you append an affix, for example, the actual pronunciation of the root/stem does not change. Consider the “-ing” suffix, whose form can be shown as [[V]-ing]. To pronounce this, you just pronounce the verb on its own and attach the suffix. This procedure does not apply to all forms of morphology. For example, take the words “parenthood” and “parental”; the respective suffixes interact with the phonology in different ways, as is evident from the variable pronunciation of the aforementioned words. In fact, the “-al” type of morphology is invisible to the phonology. In other words, the phonology reacts as if there were no morphology at all. Hence, the phonology would treat a word like “parental” in the same way it would treat a mono-morphemic word like “agenda”. So we can characterise the internal structure of a form like “parental” as:


It is this kind of morphology that is referred to as non-analytic. And since the only effect morphology can have on phonology is the presence of internal domains, it follows that this kind of morphology can have no effect on phonology, since there are no internal domains. To further accentuate the disparity between analytic and non-analytic morphology, let us look at the morpheme “un-” (which is analytic), and the morpheme “in-” (which is non-analytic).

“un-” does not change its form, regardless of what consonant follows it. This property follows from its analytic morphology. Since the prefix “n” is not adjacent to the following onset, there is no restriction in terms of phonotactics. A word like “unreal”, then, can be represented as

[[un Ø] [ri:l Ø]],

where Ø represents the empty nucleus, whereby the first of which separates the two domains. The first empty nucleus provides us with a parsing cue, along with the nucleus in the following syllable, such that we are able to perceive the prefix as a separate structure. So a string segment like “nØr” is phonologically parsable, and one can analyse the form as “un-real”.

With the prefix “in-”, the situation is different. Appending “in-” to a stem must yield a well-formed phonological domain on its own; this is because non-analytical morphology, by definition, has no internal domains. A non-analytic combination of morphemes is interpreted as follows:

φ (concat (A,B))

You concatenate the two strings and then apply phonology to the result. So a form like “in-rational” is not well-formed because “nr” is not a possible sequence. That is why the “n” has to be dropped, resulting in the form “irrational”.

So non-analytic morphology is invisible to the phonology. There are no internal domains, or any other phonological indication of morphological complexity. Non-analytic forms have the same phonological properties as any simplex form. It is reasonable to assume, then, that any form manifesting non-analytic morphology is listed in the lexicon.

Irregular forms that exhibit morphological complexity are not phonologically parsable. This is because there are no phonological hints of their complex morphological structure. It is for this reason that irregular morphology is always non-analytic.

So with regards to the kind of compounds we are looking at, compounds can be classified into four groups, viz.:

Group 1 – Analytic, singular (example, tree-eater)
Group 2 – Analytic, plural (example, rats-eater)
Group 3 – Non-analytic, singular (example, linguistics-programme)
Group 4 – Non-analytic, plural (example, mice-eater)

In the groups above, ‘analytic’ refers to the compound itself, and ‘singular/plural’ refers to the modifying noun.

The relevance of his thesis is quite evident in that it may explain why the counter-examples to Gordon’s experiment are only pseudo-counter-examples. German, for example, does make use of compound words that contain a plural form. This seems, prima facie, a blatant refutation of Gordon’s claim. However, Kaye offers us some novel way to save the Gordon-hypothesis from refutation. My point here, however, is that Gordon’s argument as it stands is untenable, but can be saved in light of Kaye’s explanation, as the plural compound words have been shown to belong to our “Group 4” type words. However, German is peculiar in that the language has more irregular forms than regular ones, which leads one to the rather quaint conclusion that the ‘irregular’ plural seems to be more regular than the ‘regular’ plural inflexional suffix. Hence, the explanation for this phenomenon has nothing to do with genetic inheritance, but can be explained by appealing to the logic of word structure. Whether that is innate is a separate question, and Gordon would have to explain why languages do it differently. One could easily do so with a parameters-like explanation, but as I will explain later, this kind of approach is unscientific, and therefore vacuous. If there are other explanations that explain the same data more consistently and more logically, why should we accept the claim that this is unequivocally a result of innate constraints? For the Nativists, they insist that innateness is the only logical explanation for the phenomenon Gordon refers to; Kaye has shown otherwise. This concludes my discussion of Kaye.

A form like “mice-eater” is acceptable because the phonology is blind to its morphological complexity. A form may also not be irregular, and still be an acceptable part of the compound, as in the plural German compound words found in the data above. As Kaye has established, complex forms with no internal domains are treated by the phonology as if they were simplex. If German can be shown to exhibit non-analytic morphology with regards to using plural modifying nouns, they do not constitute true counter-examples; or rather, Gordon’s original hypothesis can be easily adapted to accommodate these facts without actually compromising the crux of his original thesis, which could now state that the restriction applies only to constituents that exhibit analytic morphology. For example, in a compound like rattenfresser, even though ratten is the (regular) plural of ratte, it actually is non-analytic, which means that the phonology is blind to the morphological complexity of the modifying noun.

On the Origin of Language

Chomsky is of the opinion that we would have to remain agnostic as to how language came in to being. In fact, Chomsky goes so far as to insist that we can “tell the story as we like”, and that Darwinian natural selection will not suffice as an explanation since there are no precedents anywhere in the natural world for language. Language, according to Chomsky, lacks the messiness we would expect of an accumulation of accidents made good by evolutionary ‘tinkering’. Characterised by beauty bordering on perfection, it cannot have evolved in the normal biological way. Pinker disagrees with this claim. In paper with Bloom (published in 1990) as well as in The Language Instinct, it is argued that natural selection is the only viable explanation for the origin of language. Pinker draws the analogy between language and spider webs; Pinker & Bloom say that language “is a topic like echolocation in bats or stereopsis in monkeys”.

Pinker hypothesises that there must have been a mutation of some kind in our genetic make-up, which happened to be useful to that particular individual. This trait would have been passed on, as natural selection retains traits which maximise the chances of survival, therefore the grammar gene would have been spread to become what it is today. Pinker is scanty on the details, and he himself concedes that this is speculative, but goes on say that the survival benefits are very obvious, and therefore it is quite plausible for a scenario of this kind to take place. How else would humans have worked together in groups to hunt wild-life far more powerful than themselves?

The hypothesis that grammar is historically the exclusive product of a special-purpose genetically provided mental organ for grammar rests upon a negative argument: we have not managed to explain how grammar arises through general processes. Hence, we must postulate the existence of a special genetic code that is entirely responsible for building in the brain a UG organ, which will someday be located and understood in the way the retina or the heart has been located and understood.

Linguistic Relativity

We are introduced on p. 57 of the THE LANGUAGE INSTINCT to "the famous Sapir-Whorf hypothesis of linguistic determinism", named after Edward Sapir and Benjamin Lee Whorf. According to Pinker, this is the notion that language determines the way we think, and the way in which we perceive the world. He illustrates this notion with the misunderstood statement that Eskimos have hundreds of words for snow, and therefore they somehow perceive snow differently to how English speakers would. This does not make sense, since it is actually not true that Eskimos have so many words for snow, and this does not take into consideration the fact that English has over a dozen words for snow. Clearly, it does not make sense to stipulate that Eskimos have different perceptual mechanisms just because they have a different language. He also mentions that just because some languages have less colour terms than, say, English, it does not follow that speakers of these languages do not perceive colour in the same way we do. The point Pinker is making here is simply that all human beings have the same cognitive faculties, and that we would all perceive the world in the same way, even though our linguistic conventions in naming that world may differ. In fact, Pinker notes, no matter how influential language may be, the idea that language can “reach down into the retina and rewire the ganglion cells” is preposterous.

ª The bracketed “s” is referred to as a “fugen-s ” [gapping/bridging s], so-called because it is there to serve as a bridge between the two constituents, and does not in itself carry any semantic significance.

No comments: