Philosophy of Linguistics

1

 

Chapter 35

Philosophy of linguistics

 

Esa Itkonen 

 

1. Introduction 

A. Purpose 

Katz (1985) is a decent attempt to establish philosophy of linguistics as a field of its own. It

seems eminently reasonable that this field should be “conceived as a branch of philosophy

parallel to the philosophy of mathematics, the philosophy of logic, and the philosophy of

physics” (p. 1). What is less recommendable, however, is the openly expressed wish to single

out those issues that might be of interest to philosophers at large. There is no need to convince

any particular part of the general audience that philosophy of linguistics is valuable. Just like

the philosophy of any other academic discipline, it possesses an intrinsic value. 

B.  Definition 

The philosophy of linguistics is indistinguishable from the methodology of linguistics.

Linguistics is not, however, a monolithic whole, but comprises several subdisciplines, which

may overlap to a higher or lower degree: grammatical theory, typology, diachronic linguistics,

sociolinguistics, psycholinguistics, and so on. The methodology of each of these

subdisciplines has an equal right to be taken into account. To be sure, the greatest amount of

philosophical reflection has been devoted to grammatical theory.  

 Philosophy of linguistics must not be confused with philosophy of language, as expounded

e.g. in Blackburn (1984). The two overlap in semantics. To philosophy of linguistics,

however, phonology is just as important as semantics, whereas philosophy of language (as

this term is currently used) has no interest in phonology. Moreover, if in the domain of

philosophy an analogue has to be found for philosophy of linguistics, it is not philosophy of

language but philosophy of philosophy, or metaphilosophy (cf. Pap 1958, Cohen 1986,

Wedgwood 2007). 

2

 

C. A question of authorship 

It goes without saying that history of linguistics concentrates on the leading representatives of

each period. Thus, to put it bluntly, the history of linguistics deals with the best linguists.

(This entails that any purportedly value-free approach is doomed to failure.) It may seem selfevident

that these very

same

personages will

reappear

in the history

(or, rather,

historiography)

of the philosophy of linguistics. But

this cannot be quite

right. ‘Philosophy of

subject

matter

X’ presupposes methodological

self-awareness

of what it means to study X

and,

in the historical context, some

written record

of this self-awareness. This means

that the

best

linguists need not be the best philosophers of

linguistics, for two reasons. First, one may

have

left no written record of one’s methodology.

Second, even where some

record exists, it

is

not unusual for one to be a good linguist but

a bad philosopher of linguistics.    

 It follows that in that type of history that will be dealt with below, the authorship of ideas

is less clear-cut than in the history of linguistics. The ‘philosophy of person Y’ may have been

formulated by Y him-/herself; or it may have been extracted from his/her writings by

somebody else.  

D. Choosing the focus 

There have been significant linguistic traditions outside of the West, especially in India and in

Arabia (understood as a cultural rather than geographical area). For reasons of space, these

other traditions will not be treated here in their own right, but some of their contributions will

be briefly mentioned in a proper context. For some 2000 years, synchronic grammatical

analysis constituted the sole type of linguistic inquiry in the West. Other linguistic

subdisciplines, although just as important methodologically, will therefore be treated more

succinctly or not at all.  

E. First beginnings 

For a long time, the three traditions mentioned above have been monolingual, which means

that they have not been brought into existence by awareness of cross-linguistic differences (cf.

Auroux 1989: 25–26). Rather, the original motivation for linguistic inquiry has been exegesis

of some canonical text, even if grammar-writing later established itself as an independent

form of research (cf. van Bekkum et al. 1997: 287). It seems natural to think that the same

hermeneutical interest may continue to imbue even the modern-day linguistics to some extent.

To be sure, in the West another impetus for linguistic inquiry came from general philosophy

(see below). 

3

 The invention of writing has often been cited as the principal facilitating factor that 

enabled linguistics to come into being. This view, inspired by Goody (1977), certainly sounds

plausible, but it overlooks some important facts. Pāṇini composed his grammar of Sanskrit

around 400 BC, but there are no written records of either Sanskrit or its descendant Prakrit

before the edicts of the emperor Asoka (c. 250 BC). By comparison, the (Harappan) language

of the Indus valley civilization (c. 2200–1800 BC) has been preserved in numerous (albeit

brief) documents carved on stone, metal, or bone. “It is unheard of that any people having a

script never use it on hard materials” (Masica 1991: 134). It follows that Pāṇini had no written

medium at his disposal, a view confirmed by a visiting Greek who flatly stated c. 280 that the

Indians have no knowledge of written letters (op. cit., p. 135). 

 It is customary to claim that Pāṇini could not have achieved his grammar without the aid of

writing, but this is just another instance of Eurocentrism. “Inference of the existence of

writing from the feats of textual preservation and analysis accomplished by the ancient

Aryans does an injustice to the remarkable mnemonic powers developed by them” (ibidem).

Some of the mnemonic techniques are discussed in Rubin (1995). The over-emphasis put on

literacy tends to conceal what the human mind is capable of (cf. Itkonen 1991: 13–14). 

2. Antiquity 

In his dialogue Cratylus Plato (428–348 BC) raises the question concerning the nature of

language: does it exist nomōi or phusei? These are dative forms of nomos and phusis,

respectively. The former means ‘law’ and ‘convention’ while the latter means ‘nature’ in the

sense of ‘essence’. If language exists nomōi, it is conventional or arbitrary. If it exists phusei,

it is an instrument that has been devised by a mythical or imaginary ‘name-giver’

(onomathetēs or nomothetēs) so as to reveal the essence of its referent; and this it can do

insofar as it is a picture of the latter.    

 The actual examples that are marshalled to support the phusei view are rather

unconvincing, as Plato (or his mouthpiece Socrates) is himself willing to admit. Still, he

claims that this is what language would be like in an ideal world: “For I believe that if we

could always, or almost always, use likenesses, which are perfectly apppropriate, this would

be the most perfect state of language, as the opposite is the most imperfect” (435C). 

 Here the central question is whether or not there is something like ‘correctness of of

names’ (orthotēs onomatōn). It may be of some interest to note that the same question lies at

the heart of the most influential philosophical doctrine in China, namely that of Confucius: 

4

whether or not it is possible to (re-)establish the situation consonant with cheng ming, 

generally translated as ‘rectification of names’ (cf. Fung Yu-lan 1952: 60).

 In his dialogue Sophist Plato formulates the first rudiments of sentence analysis. He notes 

that (correct) statements are not just strings of words like *Walks runs sleeps or *Lion stag

horse. Rather, when one is uttering a (correct) statement of the “simplest and shortest possible

kind” like A man understands, one is “putting a thing together with an action by means of a

name and a verb” (suntheis prāgma prāxei di’ onomatos kai rhēmatos). As a consequence,

“just as some things fit together, some do not, so with the signs of speech; some do not fit, but

those that do fit make a statement” (262D–E). 

 Aristotle (384–322 BC) does not investigate language for its own sake but rather as part of

his logic, rhetoric, or poetics. Here the emphasis will be placed on the connection with logic. 

 Aristotle identifies logic with syllogistic. A syllogism is an inference with two premises

and one conclusion. The premises and the conclusions are sentences with the core structure X

is Y (cf. below). Both X and Y express (lexical) meanings ultimately definable in terms of ten

categories. Thus, preliminary to his logic proper, expounded in Analytica Priora (to use the

Latin name), Aristotle had to write the books on categories (= Categoriae) and sentences (=

De Interpretatione). As documented by Arens (1984), the two or three opening pages of the

latter book have had a lasting influence on the Western linguistic tradition. 

 The following three relations of signification are stated to exist: written language

(grammata) → spoken language (phōnai) → mental experiences (pathēmata or noēmata)  →

things (prāgmata). The link between language and mental experiences is conventional (kata

sunth

ēkēn, synonymous with nomōi). By contrast, mental experiences are pictures

(homoiōmata) of things, which means that the two must be linked by a phusei-type

connection. Unlike the linguistic expressions, the mental experiences and the things are

common to the mankind. Even if the reality (as constituted by things) is always

conceptualized or ‘mind-penetrated’, mind and reality must nevertheless be postulated as two

distinct realms. There has to be mind because there clearly are some mental experiences with

no extramental counterparts. But there also has to be reality: “that the substrata which cause

the sensation should not exist even apart from sensation is impossible” (Metaphysica 1010b,

30).      

 There are different types of sentences, of which statements, i.e. those admitting of truth or

falsity, are singled out as those needed in syllogistic. A statement has a subject (onoma, also

‘noun’) and a predicate (rhēma, also ‘verb’); as we have seen, these terms were already used 

5

by Plato. A statement is given both a semantic and an ‘actionist’ interpretation: The subject 

both signifies a substance and is that about which the predicate asserts something. The

predicate is thought to contain an either implicit or explicit copula, which means that all

statements are reduced to the canonical form X is (not) Y required by syllogistic. This entails

that all types of predicates must, rather unnaturally, be reinterpreted as one-place predicates:

e.g. Socrates loves Plato → Socrates is Plato-loving, Socrates gives books to Plato →

Socrates is books-to-Plato-giving. The copula has no lexical meaning but only a grammatical

meaning: it does not signify (sēmainein) but only consignifies (prossēmainein).  An

affirmative statement combines the meanings of X and Y, while a negative statement

separates them. 

 As explained in more detail by Kneale and Kneale (1962: II, 5–6), syllogistic deals with

four distinct types of sentences, traditionally designated by the following letters: 

 A = Every S is P

    E = No S is P

 I = Some S is P

 O  = Some S is not P

 These are the ‘general’ sentence-types, of which A and E qualify as ‘universal’ while I and 

O qualify as ‘particular’. A and O, on the one hand, and E and I, on the other, are

contradictories while A and E are contraries. E and I are ‘convertible’: No S is P = No P is S

and Some S is P = Some P is S. S and P can be understood as standing for ‘subject’ and

‘predicate’, respectively. This choice of letters anticipates the conclusion of the syllogism,

which  has the form S – P. In addition to the conclusion, a syllogism contains two premises

such that one states a relation between S and a ‘middle term’ M and the other states a relation

between M and P.  

 This issue is slightly complicated by a discrepancy between Aristotle’s original

formulations and their modern translations. In general, Aristotle uses a schema of the

following type: “If P is predicated of every M and M is predicated of every S, then it is

necessary that P should be predicated of every S.” This particular schema may be exemplified

by the following syllogism: 

  ‘Patriot’ is predicated of every maniac

  ‘Maniac’ is predicated of every soldier 

6

  -----------------------------------------------

  ‘Patriot’ is predicated of every soldier

 Furthermore, it is natural to summarize this syllogism in the following way:

  P – M

  M – S

  --------

  P – S 

 This indeed illustrates Aristotle’s original way of summarizing syllogisms (cf. Kneale and

Kneale 1962: 68). But if the premises and the conclusions are assumed to be copula sentences

of the type X is Y, then both the subject~ predicate order and the order of the premises should

be reversed: 

  Every soldier is a maniac   S – M

  Every maniac is a patriot   M – P

  -------------------------------                               -------

  Every soldier is a patriot   S – P

 For a modern reader, this difference in presentation may be a needless obstacle to an 

adequate understanding of syllogistic.

 There are four, and only four, possible ways, called ‘figures’, in which S, M, and P may 

relate to one another in a syllogism. (Aristotle himself recognized only three figures.) The

traditional presentation of these figures is something of a compromise: the order of letters has

been changed from what it was in Aristotle’s original formulation, but the order of the

premises has been retained (cf. Reichenbach 1947: 200). Evans (1982: 77), for instance, has

changed also the order of the premises. This is the new, non-traditional form of presentation: 

  1)  2)  3)  4)

  S – M  S – M  M – S  M – S 

  M – P  P – M  M – P  P – M  

   --------   -------   --------  -------    

  S – P  S – P  S – P  S – P 

7

 The syllogism given above is an instance of figure 1). Moreover, each line of a syllogism 

must exemplify one of the four alternatives A, E, I, and O. A given constellation of these

alternatives is called a ‘mood’. For instance, our example has the mood AAA (with the

traditional label Barbara); hence, its figure and mood label is 1-AAA or SAM and MAP and

SAP. Each figure contains 4 x 4 x 4 = 64 possible moods. Because there are 4 figures, there

are 256 possible moods in all. Originally, Aristotle recognized only 14 moods that are valid in

the sense that it is necessarily the case that, if the premises are true, the conclusion is true

(Kneale and Kneale 1962: 72–73). Valid conclusions of the form P – S are also possible, but

they are not part of the Aristotelian syllogistic. 

 It is a characteristic of figure 1) that its four valid moods produce, exactly, four

conclusions of the A, E, I, and O type. As we just saw, the Barbara mood produces an A

conclusion. The other three moods are as follows: If every S is M, and no M is P, then no S is

P (1-AEE, traditionally called Celarent); If some S is M, and every M is P, then some S is P

(1-IAI); If some S is M, and no M is P, then some S is not P (1-IEO). Aristotle regarded only

these valid moods of figure 1) as “perfect” or intuitively self-evident. The validity of moods

in other figures may indeed be quite difficult to grasp, for instance: If some S is M, and no P is

M, then some S is not P (2-IEO); If every M is S, and every M is P, then some S is P (3-AAI);

If some M is S, and no P is M, then some S is not P (4-IEO) (cf. Johnson-Laird 1983: 102–

103). 

 A syllogism is demonstrated to be valid by being derived from (or reduced to) Barbara or

Celarent. More informally, several generalizations have been stated about the premises of a

valid syllogism, for instance: i) If one premise is particular, the conclusion is particular. ii) If

both premises are particular, there is no valid conclusion. iii) If both premises are affirmative,

the conclusion is affirmative. iv) If one premise is negative, the conclusion is negative. v) If

both premises are negative, there is no valid conclusion.  

 Aristotle’s syllogistic is a logic of classes or, alternatively, of one-place predicates: “That

one term should be included in another as in a whole is the same as for the other to be

predicated of all of the first” (Analytic Priora 24b,25). The main difference vis-à-vis modern

formal logic concerns the interpretation of A-sentences. For Aristotle, their truth presupposes

the existence of some S, whereas the comparable universal sentence of modern logic,

formalized as x(Sx → Px), has no such “existential import”. Assuming that there are no

ghosts, the sentence All ghosts are friendly is false for Aristotle and true for modern logic. It

is only the existential import that justifies the validity of a syllogism such as 3-AAI above. 

8

The existential import also justifies the “partial conversion” Every S is P > Some P is S (via 

Some S is P). 

 Aristotle’s authority was pervasive and long-lasting. Writing in 1787, Immanuel Kant still 

claimed that in logical theory no progress at all has been achieved during 2000 years because

Aristotelian syllogistic is “complete and perfect” [geschlossen und vollendet]. From today’s

perspective, however, this kind of logic seems arbitrary insofar as it lacks its natural

foundation, i.e. propositional logic, and ignores the existence of relations. Even as an analysis

of its chosen topic, it has been characterized as “unnecessarily complicated and unelegant”

(Reichenbach 1947: 206). Yet it has its own justification: “For many centuries all the relations

asserted by Aristotle were accepted without much questioning by the thousands who studied

his work. It is therefore likely that his work is a faithful reflection of the normal usage for

sentences constructed with words like ‘every’ and ‘some’ ” (Kneale and Kneale 1962: 59).

Hence, little needs to be changed in the way that Aristotle described negation and

quantification in natural language.   

 Aristotle’s Metaphysics contains a ‘Philosophical Lexicon’ where e.g. the following

conceptual distinctions are defined: one vs many, same vs different, quantity vs quality,

necessary vs possible vs imposssible, prior vs posterior, active vs passive, part vs whole. It is

not difficult to see that these overlap with grammatical meanings that every language of the

world has to express in one way or another. 

 In sum, Allan (2010: 58) seems eminently justified to say of Aristotle that “his legacy is

overwhelming”. 

 After Plato and Aristotle, the Stoics constructed an ambitious theory that was supposed to

encompass the entire universe. It has three components: ethics, ‘physics’ (including

psychology), and logic. Logic is in turn subdivided into dialectic and rhetoric. Within

dialectic it is possible to further distinguish between grammatical analysis and the theory of

valid inference. Unlike Aristotle, the Stoics investigated language for its own sake and not

just as a precondition for doing logic. 

 In general, the Western tradition assumes that mental experiences are the meanings of

linguistic forms, ascribing — rightly or wrongly — this view to Aristotle.  The Stoic

metaphysics is strongly materialistic in the sense that as many entities as possible (inluding

qualities and relations) are interpreted as being of corporeal nature. It is therefore only the

more remarkable that, according to the Stoics, linguistic meanings (sēmainomena or lekta) are 

9

incorporeal and thus clearly distinct from the corresponding mental experiences (phantasiai), 

which belong to ‘physics’. Of course, linguistic forms (sēmainonta) too are part of ‘physics’. 

 Distinct types of sentences are recognized insofar as they express different speech acts, for 

instance, statements, questions, and requests. A verb like ‘teaches’ is an incomplete predicate

(and an incomplete lekton); a verb phrase like ‘teaches Dion’ is a complete predicate but an

incomplete lekton. A statement like ‘Plato teaches Dion’ is a complete lekton. A negation plus

a statement forms a statement. Likewise, two statements conjoined form a statement. Because

statements qua lekta are true or false while this cannot be said of sentence-meanings, it is only

as a first approximation that lekta have been equated here with meanings. It would be more

accurate to say that to a declarative sentence-form there corresponds, as its lekton, that which

it asserts to be the case.     

 Both verb and noun morphology have become objects of systematic analysis in the Stoic

grammar. It influenced both Varro and Apollonius (see below). 

 Chrysippus (282–206 BC), the leading Stoic thinker, constructed a propositional logic,

which Aristotle had failed to do. He assumed five inference schemas as basic. For simplicity,

they will be presented here with the modern notation: 

  1)   2)  3)   4)  5)

  p → q  p → q  ~(p  q)  p  q  p  q

  p  ~q  p   p  ~p

  --------   -------  -----------  ---------  --------

    q  ~p  ~q   ~q  q

 Here p → q = ‘if p, then q’, ~q = ‘not-q’, p  q = ‘p and q’. The schemas 1) and 2) are 

known as Modus Ponens and Modus Tollens, respectively. Taken together, the schemas 4)

and 5) show that the disjunction ‘or’ (= ) is not inclusive, as in modern formal logic, but

exclusive: ‘p or q but not both’. Of course, the inferences 4) and 5) are also valid if the second

premise is either q or ~q, with ~p and p as the respective conclusions.  In the schema 3) the

negation (= ~) and the conjunction (= ) combine to define an inclusive disjunction (‘not-p or

not-q or both’). The inference is valid if the second premise is either p

 or q, with ~q and ~p as

the respective conclusions. But if the second premise is ~p, neither q nor ~q follows, because

the truth of both is compatible with the truth of the premises; and the same applies, mutatis 

10

mutandis, to ~q as the second premise.  The law of the excluded middle, i.e. p  ~p, which is 

abandoned by some schools of logic, is unquestioningly endorsed by the Stoics.

 The five basic or ‘indemonstrable’ schemas are used by Chrysippus and his followers as 

starting points from which an indefinite number of other valid schemas or sentences could be

derived as theorems. (The Stoics were perfectly well aware that inference schemas can be

transformed into conditional sentence-schemas, e.g. Modus Ponens into [(p → q)   p ] → q.)

Thus, what they were trying to do was construct axiomatic (propositional) logic. On the basis

of the extant evidence it is difficult to appreciate the extent to which they actually carried out

this program. At least the following theorems were proved (cf. Kneale and Kneale 1962: 165–

169). 

(i)  If the first, then if the first, then the second; but the first; therefore the second = by two

applications of schema 1): [p → (p → q)]  p yield (p → q), and (p → q)  p yield q.  

(ii)  If the first and the second, then the third; but not the third; on the other hand the first;

therefore not the second = by one application of schema 2 and one application of schema 3: (p

 q) → r and ~r yield ~(p  q), and ~(p  q) and p yield ~q. 

(iii)  Either the first or the second or the third; but not the first; and not the second;

therefore the third = by two applications of schema 5:

[p  r)] and ~p yield (q  r), and (q 

r) and ~q yield r.  

(iv) If the first, then the first; but the first; therefore the first = by one application of the

schema 1: p →p and p yield p. 

(v)  Either the first or not the first; but the first; therefore not not the second = assuming

the law of the excluded middle and applying schema 4: p  ~p and p yield ~~p.  

(vi)  Either the first or not the first; but not not the first; therefore the first = assuming the

law of the excluded middle and applying the schema 5: p  ~p and ~~p yield p. 

 Taken together, theorems (v) and (vi) establish the equivalence p  ~~p. Theorem (iv) is

important because it shows that the same methods must be applied to prove any  non-basic

schema, regardless of whether it is felt to be intuitively non-obvious or obvious. This is

something that the critics of Stoic logic failed to understand. Apparently also the following

theorem was proved: [(p → q)  (p → ~q)] → ~p. This is the all-important inference schema

known as reductio ad absurdum, which says that a sentence is false if it contains a

contradiction. 

11

 Although many aspects of Stoic logic survived in fragmentary comments, the basic insight, 

i.e. the invention of propositional logic, fell into oblivion. It was rediscovered only in the 20th

century, thanks to Łukasiewics (1935). Thus, it is part of the “hidden history” of the Western

thought. While Stoic logic remained without genuine influence, Euclid’s Elements provided

the generally acknowledged model for axiomatic thinking until the end of the 19th century. It

is interesting to note that Pāṇini’s grammar too exemplifies the idea of axiomatics (cf. Itkonen

1991: Ch. 2, esp. pp. 38–44). Comparable to the Elements, it was for more than 2000 years

the cornerstone of higher education in India.    

 After Plato and Aristotle, linguistic inquiry received two types of impetus. One came from

Stoic philosophy. Another came from Alexandrian philology, starting in the third century BC.

Marcus Terentius Varro (116–27 BC) was influenced by both lines of thinking. The aspect of

Varro’s methodology that will be singled out here is the notion of explanation in his book De

Lingua Latina. 

 This is the crucial passage: “The origins of words are therefore two in number, and no

more: impositio and declinatio; the one is as it were the spring, the other the brook. Men have

wished that imposed nouns should be as few as possible ... but derivative nouns (nomina

declinata) they have wished to be as numerous as possible” (VIII,5).  

 Impositio designates the original act of name-giving, and Varro refers to “those who first

imposed names upon things” (“illi qui primi nomina imposuerunt rebus”) (VII,7). This is

Varro’s counterpart to Plato’s nomothetēs, with the difference that the Varro-type name-giver

is an ordinary, occasionally fallible human being. Declinatio designates the process by which

new words are produced from the original simple words. In practice it covers inflection and

derivation. 

 If declinatio is applied to e.g. 1000 words resulting from impositio, millions of new words

can easily be shown to result (VI,36–39). First of all, it may be admitted that the original 1000

words are unexplainable. Still, it is no mean feat for the grammarian to have explained such a

large number of the new words. The term for ‘explaining’ is here either ostendere or expedire. 

 But secondly, this is not all that the grammarian can do. The very disproportion between

few imposed words and many derived words is explained by the fact that it imposes a

minimum load to memory. Such and similar functional explanations are ultimately based on

the instrumental character of language: “I grant that speech has been produced for utility’s

sake” (“Ego utilitatis causa orationem factam concedo”) (IX, 48; also VIII, 30). Now,

language would be a very bad instrument if every word had to be learned separately. 

12

     Third, we need not admit that impositio operates in a totally arbitrary way: “nature was 

the man’s guide to the imposition of names” (VI, 3). On the one hand, the lack of plural forms

in mass-nouns is ontologically motivated (IX, 66). On the other, the lack of gender

distinctions in the names of non-domestic (and thus less important) animals is functionally

motivated (IX, 56). Varro is here outlining a moderate version of the phusei view. 

 Much of the discussion is carried out in terms of the ‘analogy vs anomaly’ dichotomy. For

instance, inflection is correctly asserted to be more analogical and less anomalous than

derivation. A phenomenon is explained by exhibiting its place in a larger analogical structure.

For instance, Varro (X, 47–48) gives the following account of the Latin tense/aspect system: 

 INFECTUM      PERFECTUM

 Imperfect  Present    Pluperfect  Perfect

 legebam  lego    legeram  legi

 ----------- = ----------  = ------------- = ----------

 lego   legam    legi   legero

 Present  I Future   Perfect   II Future

 The generalization implicit in the data is presented here by means of a complex analogy. 

On both sides, the first-level analogy is temporal: the relation of past to the present is the

same as the relation of the present to the future. The two sides of each analogy share a

common member (= present), which means that each is an analogia coniuncta. Moreover,

both INFECTUM and PERFECTUM contain the same (temporal) analogy, which means that

they are analogous to each other at the second level. They have no common members, which

means that each is an analogia disiuncta. This second-level analogy is aspectual. The only

weakness of this elegant description is that the aorist function of the perfect remains

unaccounted for. 

 The Varro-type analogy is not just language-internal, but extends — phusei-like — also to

the relation between language and extralinguistic reality: “The basis of all regularity

(analogia) is a certain likeness, that, as I have said, which is wont to be in things and in

spoken words and in both” (X, 72; also IX, 63). Finally, Varro (IX, 23) justifies analogy in

language by referring to the ubiquity of analogy in general: Quae enim est pars mundi quae

non innumerabiles habeat analogias? (‘For what part of the world is there which does not

have countless regularities?’) 

13

 The oldest extant sentence-analysis in the Western tradition is given in Peri suntaxeōs by 

Apollonius Dyscolus (c. 80–160). This title should not be translated as ‘syntax’, because the

book deals equally with the meaning and the form of sentences.  

 Apollonius exhibits considerable freedom in inventing and manipulating his data. Consider

the following example (I, 14): 

 ho   autos   anthrōpos   olisthēsas   sēmeron   kata-epesen

  art   pron    noun           participle   adv           prep-verb

 The same    man            slipping    today        down-fell

 This sentence has been so constructed as to contain all word-classes except the 

conjunction. The idea of name-giving (thesis tou onomatos) still survives, in the sense that the

word classes are supposed to have been invented in the order of their importance: first the

noun, then the verb, and so on. Now, it is of course Apollonius’ own linguistic intuition that

assures him of the correctness of this self-invented exmple sentence. Next, the words of this

sentence are deleted one by one, on the correct assumption that only those words whose

deletion results in incorrectness are necessary to sentencehood; and, again, the only criterion

of incorrectness is Apollonius’ own intuition. In this way the original sentence is reduced step

by step to anthrōpos epesen. Looking back, we realize that it was Plato’s intuition, and

nothing else, that distinguishes the correct The man understands from the incorrect *Walks

runs sleeps. 

 All this is self-evident if the data of grammatical analysis is taken to be the normal

linguistic usage which is by definition accessible to intuition. The grammarians of antiquity,

however, view the data in more complex terms. Varro distinguishes between usage

(consuetudo) and reason (ratio), even if he admits that they often coincide: “Therefore as each

one ought to correct his own usage if it is bad, so should the people correct its usage. ... [T]he

people ought to obey reason, and we individuals ought to obey the people” (IX, 6). Hence a

nation should correct its language if it is contrary to ratio, but an individual has no choice but

to follow the consuetudo of his/her nation. 

 Apollonius echoes the same view: “I rely not merely on poetical citation ...., but on

common everyday usage, the practice of best prose-writers, and most of all, on the force of

theory” (II, 49; emphasis added). The resulting attitude vis- -vis the data is somewhat

“warped”. One is forced to accept the usage as it is, but then one counteracts by trying to find

a rational justification for it even where none exists in fact. “This notion that everything in 

14

grammar must have a reason, that nothing is arbitrary or random, pervades A.D.’s work” 

(Householder 1981: 43).

 As far as Apollonius’ methodology is concerned, this is the crucial passage: “So we’d 

better stop and explain what the actual cause of the ungrammaticality is, not by mere citation

of examples as some linguists do, pointing out the ungrammaticality without explaining its

cause. But if you don’t grasp the cause, it is an exercise in futility to cite examples” (III, 6).

The central explanatory notion is akolouthia (or katallēlothēs), translated as ‘concordance’. It

subsumes agreement and government, but is not restricted to these two. Examples are given

throughout the book, the most revealing ones in III, 22–34. This is the general constraint: “No

part of speech [combined with another] can be ungrammatical with respect to a category

which it fails to distinguish” (III, 51). Looking back, we realize that Plato’s distinction

between A man understands and *Walks runs sleeps is based on pretheoretical use of 

akolouthia. 

 Apollonius makes extensive use of the descriptive apparatus ‘deep structure –

transformation – surface structure’, connecting e.g. the following types of sentences: N V1-ed

and (then) he V2-ed, N V1-ed and V2-ed, N who V1-ed V2-ed, V1-ing N V2-ed. He also

postulates uniform performative deep structures for sentences expressing different speech

acts. This just goes to show that the history of linguistics is shorter than one might have been

led to think. The very ontology-cum-methodology of grammatical analysis must possess some

peculiar feature that makes this possible. 

 Because the Hellenistic Greek utilized by Apollonius is fully comprehensible only to a

small group of experts, Peri suntaxeōs remained practically unknown until Householder’s

1981 translation. Therefore, as surprising as it may sound, no adequate understanding of the

Western linguistic tradition was possible, in practice, before 1981.  

3. From the Middle Ages to the End of the 18th Century  

The scholastic grammarians, called Modistae (c. 1250–1320), constructed a universal

grammar (grammatica universalis) which had the following general structure. First, there are

the ontological categories of thing, action, and quality, called the “modes of being”. Second,

there are the corresponding mental concepts, called the “modes of understanding”. Third,

there are the corresponding word classes, i.e. noun, verb, and adjective, called the “modes of

signifying”. There is a general iconicity between the three levels, but none of them is

redundant: “Although things cannot be understood apart from the modes of understanding, the 

15

human mind makes nevertheless a distinction between the things, on the one hand, and the 

modes of understanding, on the other” (Boethius de Dacia 1980 [c. 1280]: 70). Considering

the iconic relation between thing and noun, on the one hand, and action and verb, on the

other, we now see that Plato would have been able to build a more convincing case for the

phusei view, if he had been speaking about word classes instead of individual words. 

 According to Aristotle, both the reality and the mind are universally the same whereas

languages which express what’s in the mind vary arbitrarily. The Modistae found this

inconsistent, and correctly so. If universally valid distinctions are expressed by the various

languages, these too must possess some universal core which is just hidden behind superficial

differences. 

 The Modistic universal grammar is an ambitious theory insofar as the modi significandi are

meant to be explained — via the modi intelligendi — by the modi essendi. This is languageexternal

explanation. The modi

significandi

are meant

to provide language-internal

explanations

in the sense of Apollonius-type akolouthia.

 There is a grave discrepancy between what the Modistae tried to achieve and what they

achieved in fact. Because their data basis remained restricted to Latin, they were caught in a

vicious circle: the existence of a linguistic category suggests the possible existence of a

corresponding ontological category, which is then used to ‘explain’ the former. Furthermore,

the levels of language and ontology came to be conflated in practice.  

 The same confusion hampers Modistic syntax. On the one hand, the verb is thought to

depend on the noun, because the action (qua the ontological counterpart of the verb) depends

on the thing (qua the ontological counterpart of the noun). On the other hand, there is an

attempt to separate linguistic criteria from ontological ones: “The sentence is made complete

by the word-class which governs the others and is governed by none, namely the verb”

(Sigerus de Cortraco 1977 [c. 1330]: 51). Let it be mentioned that Sibawaihi (d. c. 795), the

founder of the Arab linguistic tradition, applied the notion of dependency systematically in a

purely language-internal sense (see Owens 1988).   

 Plato and Aristotle had regarded thinking as a form of silent speaking. The Stoics assumed

the existence of a mental language which is free from the quirks of the oral language. The

Modistae took the existence of a mental language for granted. Perhaps the most precise

definition of a mental sentence (propositio mentalis) is given by William Ockham (c. 1285–

1349) in his book Summa Totius Logicae.  

16

 William assumes that there are no universal concepts like ‘man’ and ‘white’ (referred to by 

the corresponding terms S and P), but only individual men and, as it were, individual

occurrences of whiteness (referred to by the corresponding individual-names S-i and P-j).

Hence, the meaning normally expressed by This man is white ought to be expressed by an

identity-sentence like S-7 = P-29. The Aristotelean general sentences are now transformed

accordingly, in two steps. First, universally and existentially quantified sentences like Every S

is P and Some S is P are reduced to simple or non-quantified sentences, i.e. either

conjunctions S-1 is P and S-2 is P and ... or disjunctions S-1 is P or S-2 is P or ... This was

already “an interesting novelty” (Kneale and Kneale 1962: 268). Second, subject~predicate

schemas X is Y and X is not Y are replaced by identity schemas X = Y and X ≠Y. As a result,

the general sentences are given the following truth-conditional definitions (cf. Loux 1974:

28–31): 

A) Every S is P   [(S-1 = P-1)  (S-1 = P-2)  ...]  [(S-2 = P-1)  (S-2 = P-2)  ...]  ...   

E) No S is P   [(S-1 ≠ P-1)  (S-1 ≠ P-2)  ...]  [(S-2 ≠ P-2)  (S-2 ≠ P-2)  ...]  ...

I) Some S is P   [(S-1 = P-1)  (S-1 = P-2)  ...]  [(S-2 = P-1)

 (S-2 = P-2  ...]  ...

O) Some S is not P   [(S-1 ≠ P-1)  (S-1 ≠ P-2)  ...]  (S-2 ≠ P-2)  (S-2 ≠ P-2)  ...]  ...

 Thus, the A-sentence is reanalyzed as a (finite) conjunction of disjunctions of identity-

sentences: this S coincides with this P or that P or etc. and that S coincides with this P or that

P or etc. and etc. The E-sentence is reanalyzed as a conjunction (of conjunctions) of negated

identity-sentences. And so on. It goes without saying that this type of reanalysis (which is

supposed to be metaphysically motivated) entails a huge gain in logical simplicity. Certainly,

the analysis of sentences with proper names as subjects is less natural: asserting Socrates to be

a man amounts to asserting that Socrates is identical with Socrates or with Plato or with

Hermogenes, etc.    

 Next, sentences of increasing complexity are subjected to the same type of reanalysis. The

logical structure of a complex sentence (exponibile) is revealed by reducing it to a set of

simple sentences (exponentes) so as to ultimately reach the level of affirmative or negative

identity-sentences. Consider the sentence Socrates, insofar as he is a man, has a colour.

Mechanical rules are given to reduce it to the following more basic sentences: Socrates has a

colour; Socrates is a man; Every man has a colour; If a man exists, then something which has

a colour exists. 

17

 This type of expositio analysis was also practiced, even if less rigorously, by the Modistae. 

For instance, the sentence Sedentem ambulare est impossible ‘It is impossible for a sitting

person to walk’ was analyzed as having the following three-level (‘deep’) structure: [[[qui est

sedens] ambulat] est impossibile]. 

 In its austerity William Ockham’s metaphysics has strong affinities with the Stoic

metaphysics and differs sharply from the one generally adopted by the Modistae. Since the

existence of universal concepts is denied, it seems natural to assume that it is only by their

names that things are kept together, as it were. Hence the designation ‘nominalism’. As the

redefinitions of the A, E, I, O sentences eloquently show, logical simplicity is achieved at the

cost of extreme outward complexity. Certainly the language constructed by William could not

function as a mental language in the sense of “silent speaking”. 

 The Modistic universal grammar concentrates on ontological explanation of simple

sentences while logical linguistics as represented by William Ockham concentrates on truthfunctional

definition of complex

sentences. In spite

of their metaphysical

disagreements,

these

two

approaches should be thought of as complementing

rather than contradicting

each other.

The

linguistics of Renaissance stays far behind

what

was accomplished

in the Middle

Ages.

This

teaches an important

lesson. It is perfectly

possible that people come

to reject a theory

not

because

they have cogent

arguments

against it,

but just because they are — or think

they

are

— fed up with it. 

 The Aristotelian/Modistic heritage was exploited by all non-pedagogical grammars written

in the 16th and 17th centuries. For instance, Grammaire générale et raisonnée or the PortRoyal

Grammar

(1660) by Antoine Arnauld and Claude

Lancelot concentrates

on the first two

terms

of the Aristotelean trichotomy

‘concept

– sentence – syllogism’.

Peter Abelard had

already

postulated three mental

processes

corresponding to these three

levels (plus

a specific

process

for forming

disjunctive/conditional sentences).

In the same

vein, the Port-Royal

grammarians

and logicians (see below) postulate

the three

processes

of conceiving

(concevoir),

judging (juger),

and inferring (raisonner).

Conceiving produces an “idea”, and

judging

consists in combining

two ideas to make

an affirmation.

(Inversely, a negation entails

separating

the two ideas.) A judgment

is called a

“proposition”; what is affirmed

is called

“attribute”

while that about which it

is affirmed

is called “subject”;

the subject

is combined

with

the attribute by means

of an explicit or implicit

copula.

There

are

two

basic ontological

categories,

namely

substance and accident, with

noun and adjective/participle as their

linguistic

counterparts.     

18

 If someone says I just saw a dog, what (s)he means is that (s)he saw a certain thing, not 

that (s)he saw an idea of this thing. In his authoritative formulation of the ‘language – mind –

reality’ trichotomy in De Interpretatione, however, Aristotle clearly states that words signify

ideas (or “mental experiences”), not things; it is by ideas that things are signified. This

unfortunate formulation gave rise to misunderstandings, which are still perpetuated in the

philosophy of the 17th and 18th centuries, including the Port-Royal grammar. As against this,

William Ockham echoes Peter Abelard in asserting quite explicitly that both spoken words

and ideas (= “mental words”) signify things, even if the former do so “in subordination” to the

latter.  

 According to the Port-Royal grammarians, the complex sentence, or “figurative

construction”, Dieu invisible a créé le monde visible (‘The invisible God has created the

visible world’) is derived from the following “simple constructions”: Dieu est invisible, Dieu

a créé le monde, Le monde est visible. This is a continuation of the medieval expositio. And of

course, Apollonius Dyscolus had already noted the transformational relations between

participle constructions, relative constructions, and co-ordinate constructions. 

 The three above-mentioned mental processes are dealt with in the first three parts of La

logique, ou l’art de penser, or the Port-Royal logic (1662), by Antoine Arnauld and Pierre

Nicole. In conformity with Descartes, innate ideas are assumed to be identical with those in

God’s mind. Ideas qua mental terms may be either singular or general, their linguistic

counterparts being proper nouns and common nouns. The intension (compréhension) of a

general term X is the set of attributes which constitute it. This encompasses not just the

meaning of X but also what it entails. The extension (étendue) of X encompasses, somewhat

inconsistently, both the subtypes of X and the individuals to which X applies. The intension

and the extension of X are inversely proportional. The “quasi-mathematical” presentation of

syllogistic by Arnauld and Nicole  “continued to dominate the treatment of logic by most

philosophers for the next 200 years” (Kneale and Kneale 1962: 319–320). The fourth part of

the Port-Royal Logic, which deals with general methodology, shows the influence of

Discours de la méthode by Descartes. Special emphasis is placed on axiomatics: “[One ought]

to prove all propositions which are at all obscure, using in the proof of them only preceding

definitions, agreed axioms, and propositions already demonstrated.”      

 The tradition of the Port-Royal grammar was continued by César Chesnau Dumarsais

(1676–1757) and by Nicolas Beauzée (1717–1789), each of whom produced his own version

of general grammar. 

19

 

4. The Modern Era 

A. The Modern European Tradition  

From the beginning of the 19th century, the history of languages became the focus of

attention. Karl Ferdinand Becker tried to revitalize the tradition of general or philosophical

grammar in his Organism der Sprache (1827, 2nd ed. 1841), but his work was largely

ignored. The same is true of Anton Marty’s (1847–1914) efforts to elicit the “inner form” of

language. Similarly, Edmund Husserl’s (1859–1938) notion of a  “pure grammar”, as defined

in his Logische Untersuchungen (2nd ed. 1913), failed to have an impact on linguistics.

Parret’s (1976) interpretation of Marty and Husserl is informed by more recent developments

in linguistics.    

 Hermann Paul (1846–1921) is a central figure at this juncture insofar as his methodology

constitutes a synthesis of general linguistic theory and of diachronic linguistics. For quite

some time, most of his successors in the field of general theorizing showed little interest in

diachronic linguistics. Ultimately, however, there has been a massive de facto return to the

position held by Paul (see Section 5 below). It is not easy to summarize all these vicissitudes

into a coherent historiographical narrative, but first I will pursue the story of general linguistic

theory until the present day; second, I will take up the subsidiary story of diachronic and

typological linguistics. 

 The following account of Paul’s views is based on the Introduction and Chapters I–II of his

Prinzipien der Sprachgeschichte (’Principles of the History of Language’, first edition 1880).

Methodology (Prinzipienlehre) is the key for transcending the outdated dichotomy between

philosophy and science. Each science has its own methodology. There are two basic types of

science, namely nomothetic sciences (Gesetzeswissenschaften) and historical sciences

(Geschichtswissenschaften). Experimental physics is a typical nomothetic science.

Nomothetic science does not coincide with natural science, however, because there are also

sciences that concentrate on the history of either inorganic or organic nature (in particular,

evolutionary theory). All cultural sciences are historical sciences. Among the cultural

sciences, linguistics has achieved the most exact results, which has created the misconception

that linguistics is actually a natural science. Experimental psychology (which was just

emerging) is the most important auxiliary science of linguistics; although nomothetic in

nature, it is separated from natural sciences by the presence of the “psychic element”. 

20

 Historical grammar is scientific in the sense that it explains the causal connection 

(Kausalnexus) between successive states of language. No non-causal account can be

genuinely explanatory. A given state of language of language (Sprachzustand) is described by

a corresponding (synchronic) descriptive grammar, exemplified by Paul’s own grammar of

Middle High German (Paul and Mitzka 1960 [1881]). Because states of language are

abstractions, descriptive grammars are non-causal and prima facie non-scientific in character.

Yet they qualify as scientific in a secondary sense insofar as their existence is presupposed by

historical grammar. Any descriptive grammar remains insufficient until it is completed by a

psychological grammar of the same state of language. Although the two sets may overlap,

grammatical categories, which result from conscious reflection, must be clearly distinguished

from psychological ones, which belong to the unconscious mind. For instance, grammatical

subject and predicate are quite different from psychological subject and predicate (cf. Elffersvan

Ketel 1991: 253–256). Apart from actual non-correspondence,

the difference consists in

the

fact that

grammatical

structure

is categorical

(‘either-or’)

whereas psychological structure

exhibits

a huge number

of subtle gradations (‘more-or-less’).

Finally,

grammatical

structure

must

also be distinguished from

(purely) logical

structure. (Occasionally,

‘psychological’ and

‘logical’

are also used as synonyms.)

 The subject matter of a descriptive grammar is Sprachusus, i.e. the usage of a linguistic

community, which averages (Durchschnitt) over idiolects. Linguistics differs from other

cultural sciences insofar as there is no division of labor among its subjects apart from the

speaker vs hearer dichotomy. It is this homogeneity of the data that has made it possible for

linguistics to achieve such exact results. It is for the same reason that the usage can be

investigated in practice by concentrating on a single individual. When this is the grammarian

him/herself, careful self-observation is needed.  

 Language usage is constantly renewed in social interaction. The language of every

individual is influenced by the language of others. Unlike in biological evolution, “parents

can become children of their own children”. The contact between two minds is possible only

by means of an external physical signal sent by one person and received by the other. What

we know or assume about other minds is based on the analogy of our own mind. There is no

collective mind in addition to individual minds.  Finally, there is the central question of

linguistic methodology: how do the social Sprachusus and the linguistic activity of an

individual (individuelle Sprechtätigkeit) influence each other? 

21

 Georg von der Gabelentz (1840–1893) raises the same question in his Sprachwissenschaft 

(‘Linguistics’ 1901 [1891]), assuming a dichotomy between Einzelsprache or Sprachzustand

(‘a language’) and its concrete manifestation, i.e. Rede (‘speech’). The former is a system

whose parts are related in such an organic way that none can be changed or removed without

affecting the whole (op. cit. p. 481). 

 The same question is also raised by Ferdinand de Saussure (1857–1913) in his (1916) book

Cours de linguistique générale. For him the social aspect (langue) is clearly primary vis-à-vis

the individual aspect (parole) and constitutes therefore the genuine subject matter of

linguistics. (Remember that according to Paul the equivalent of langue is the subject matter of

descriptive grammar which remains subordinated to historical grammar.) Yet it is enough to

investigate the consciousness of individual speakers (i.e. their linguistic intuition), in order to

elicit the properties of langue.   

 Saussure uses the term langage to designate the superordinate concept which subsumes

both langue and parole. Langage in turn issues from the language capacity (faculté de

langage), which is ‘natural’, rather than social (like langue) or mental (like parole).

Moreover, since parole is characterized as an “act” (of speaking and understanding), it must

encompass the psycho-physiological performances, on the one hand, and the physical

utterances, on the other. 

 Saussure’s overall conception is narrower than Paul’s. Diachrony is of secondary

importance only; parole is clearly subordinated to langue; and there is no room for

psychological grammar (as study of the unconscious mind). 

 The “linguistic sign” is characterized as a mental entity (entité psychique) consisting of a

“concept” and an “acoustic image”, also called “signified” and “signifier”, respectively.

Langue is said to be constituted by linguistic signs, but this is inconsistent because langue is

defined as “a social institution” whereas, as we just saw, the linguistic sign is individualpsychological.

The same

contradiction

recurs in the passages where langue

is located

“in the

brain”.

 A further distinction is introduced within both components of the linguistic sign, namely

that between “substance” and “form”. The substance of the signifier is constituted by the

whole spectrum of vocal sounds while the substance of the signified is something like thought

without conceptual distinctions. By “form” is meant the way that each language structures the

two substances. The notion of form is defined more narrowly as that of value. The values, or 

22

identities, of linguistic units are determined by their mutual relations. This is the core idea of 

structuralism.

 Linguistic signs are simultaneously situated on two dimensions, namely syntagmatic and 

associative. The latter term is used in its etymological sense, i.e. as standing for the relation

between a linguistic sign X and anything that X can make the speaker think of, due to the

similarity of either form or of meaning. Thus, “associative relation” is identical with what

Paul called Verbindung zwischen Vorstellungsgruppen. 

 In his (1939) Grundzüge der Phonologie Nikolaj Sergeyevich Trubetzkoy (1890–1938)

solves the contradictions that beset Saussure’s language-conception (see the sections

“Phonologie und Phonetik”, pp. 5–17 and “Zur Definition des Phonems”, pp. 37–41).

Because langue (now called Sprachgebilde) is a social institution, its units — represented

here by the phoneme — cannot be reduced to psychological or physiological phenomena,

which belong to parole (now called Sprechakt). As normative entities, phonemes belong to

the “world of relations, functions, and values”, and not to the “world of empirical

phenomena”. “The norm ... cannot be determined by measurement and computations. ... The

system of language is beyond ‘measurement and number’. ... [In] the natural sciences there is

no equivalent for the dichotomy Sprachgebilde vs Sprechakt.” Because structure is created by

function, structuralism is functionalism. It is regrettable that Trubetzkoy’s philosophical

contribution has remained either ignored or misunderstood. 

 In his (1963 [1943]) book Prolegomena to a theory of language Louis Hjelmslev (1899–

1965) summarizes Saussure’s language-conception as the following set of dichotomies:

system vs process, expression vs content, form vs substance, syntagmatic vs paradigmatic.

These dichotomies are integrated into a closely-knit theory, and three “functions” are added

between formally defined units, namely “interdependence” (= if A, then B, and vice versa),

“determination” (= if A, then B or not-B, and if B, then A), and “constellation” (= if A, then B

or not-B, and vice versa). Hjelmslev seeks to use these three functions as the basis for

constructing a universally valid descriptive technique tantamount to a universal grammar.

First, grammars of particular languages and, second, all and only correct sentences of each

language ought to be derived from the overall theory. This approach, called “glossematics”,

seems to result in a somewhat skeletal view of language. But it is precisely for this reason, on

the other hand, that it provides a plausible framework for general semiotics. 

 Glossematics is presented as an axiomatic theory. Therefore it is natural that when

alternative descriptive techniques are evaluated, “that procedure must be chosen which 

23

ensures the simplest possible result of the description” (Spang-Hanssen 1966 [1949]: 234). 

But there is no absolute notion of simplicity. Rather, “the simplicity of a description ... is a

relative  concept, which has significance only if the purpose to which the description is

applied is indicated” (p. 235). 

B. American Structuralism and its Aftermath 

At this point we must go back to the beginning of the 20th century in order to pick up the

story of American structuralism. Franz Boas (1858–1942) edited the Handbook of American

Indian Languages, whose first volume appeared in 1911, and it laid the foundations of

anthropological linguistics. The Introduction of this book entails an interesting tension. On the

one hand, the purpose is to capture the “inner form of each language”, with the consequence

that “no attempt has been made to compare the forms of the Indian grammars with the

grammars of English, Latin or even among themselves” (Boas 1911: 81). On the other hand, it

is clear that “the occurrence of the most fundamental grammatical concepts in all languages

must be considered as proof of the unity of the fundamental psychological processes” (p. 71;

emphasis added).  

 The same tension is evident in Edward Sapir’s (1884–1939) Language, published in 1921.

In his sentence analysis (pp. 35–36) Sapir unknowingly follows the example of Apollonius.

Consider the sentence The mayor of New York is going to deliver a speech of welcome in

French. This sentence can be simplified by deleting of New York, of welcome, and in French,

but at this point the “process of reduction” must stop. For instance *Mayor is going to deliver

would be incorrect. What remains is the familiar Aristotelian dichotomy of “subject of

discourse” and “predicate”, which combine to constitute “the linguistic expression of a

proposition”. This is also the definition of “sentence”. The sentence analysis has an

ontological justification: “We must have objects, actions, qualities to talk about, and these

must have their corresponding symbols in independent words or in radical elements” (p. 93).

While “subject” and “predicate” are functional notions, the corresponding formal notions are

“noun” and “verb”, which have thing and action as their extralinguistic counterparts. Nothing

seems to have changed since antiquity. 

 However, the accumulated evidence of American Indian languages induces Sapir to boldly

assert that “no logical scheme of parts of speech ... is of the slightest interest to the linguist.

Each language has its own scheme” (p. 119). But then the following concession has to be

made immediately: “No language wholly fails to distinguish noun and verb.” 

24

 Like Hermann Paul, Sapir regards linguistics as a cultural science but, unlike Paul, he 

wishes to keep linguistics strictly separate from psychology: “We can profitably discuss the

intention, the form, and the history of speech ... as an institutional or cultural entity, leaving

the organic and psychological mechanisms back of it as something to be taken for granted” (p.

11). Let it be added that when Sapir (1933) deals with the “psychological reality of

phonemes”, what he has in mind is conformity with “phonemic/phonological intuitions”, not

some unconscious structure to be elicited by experimental psychology. 

 In his 1933 Language Leonard Bloomfield (1887–1949) shares Sapir’s view of the proper

relation between linguistics and psychology: “... we can pursue the study of language without

reference to any one psychological doctrine” (p. vii). “The findings of the linguist, who

studies the speech signal, will be all the more valuable for the psychologist if they are not

distorted by any prepossessions about psychology” (p. 32). Indeed, Hermann Paul is criticized

precisely for having tried to include psychological grammar in linguistics: “[Paul]

accompanies his statements about language with a paraphrase in terms of mental processes

which the speakers are supposed to have undergone. The only evidence for these mental

processes is the linguistic process; they add nothing to the discussion, but only obscure it” (p.

17). This criticism applies with more justification to the Aristotelian-cum-Modistic tradition

that preceded Paul than to Paul himself because, instead of just paraphrasing, he actually tried

to show that “mental processes” and “linguistic processes” do not coincide. Certainly, as he

freely admits, he was hampered in this attempt by the nascent state of experimental

psychology. 

 Bloomfield has been hugely criticized for not heeding his own advice and thus dismissing

the analysis of lexical meanings in the name of crude behavioristic psychology. This agrees

with his endorsement of ‘logical empirism’, or the prevalent philosophical doctrine of the

1930s, which advocated methodological monism, with physics as the model science. But the

criticism is not quite accurate. Being a grammarian, Bloomfield is interested in grammatical

rather than lexical meanings; and his treatment of grammatical meanings (= “[epi]sememes”)

is in no way distorted by any behaviorist prepossessions. This is just as true of an

“episememe” like “action” (p. 166) as it is of a “sememe” like “more than one thing” (p. 216).

Bloomfield’s behaviorism merely exemplifies the all-too-common gulf between one’s selfprofessed

methodology and the methodology

implicit in

one’s actual descriptive

practice.     

 In principle, Bloomfield (1933) portrays grammatical description from the perspective of a

field linguist who has to start from scratch: “Suppose we hear a speaker say John ran and a 

25

little later hear him or some other speaker say John fell. ... [I]f we are lucky, we may hear 

some one utter the form John! ... we may later hear the form Bill ran ... we may hear a form

like Dan fell” (p. 159). In practice, however, no corpus of actually uttered sentences is ever

used in this book. Rather, Bloomfield just lets his own linguistic intuition produce the data, as

he did in the passage just cited. 

    The practice of the ordinary working grammarian is sanctioned in the following way: “In

no respect are the activities of a group as rigidly standardized as in the forms of language. ...

A linguistic observer therefore can describe the speech-habits of a community without

resorting to statistics” (p. 37). The implications of this fact for the general philosophy of

science have been spelled out by Hymes and Fought (1981: 175): “It was considered a

decisive accomplishment to show the existence of qualitative structure in a sphere of human

life ... Linguistics was a demonstration of the possibility of rigorous formal analysis of a sort

not requiring sampling, statistics, or other techniques derivative of a natural science

orientation.”  

  Zellig S. Harris (1909–1992) and Charles F. Hockett (1916–2000) further develop the

structuralist approach. For Bloomfield (1933: 247–251), a “substitute” is a minimal unit (like

I or did) that can replace any member of a given form-class. Harris (1946) concentrates on

“extending the technique of substitution from single morphemes (e.g. man) to sequences of

morphemes (e.g. intense young man)”. Expressions of whatever length that pass the

“substitution test” are members of one and the same “substitution class”. Substitution classes

constitute the basis for a bottom-up sentence analysis. Wells (1947) reformulates substitution

as “expansion”. For instance, the sentence The king of England opened Parliament is an

expansion of the sentence John worked. Expansion is the inverse of, and conceptually

equivalent to, reduction as practiced by Apollonius and Sapir. Expansion constitutes the basis

for a top-down sentence analysis in terms of immediate constituents (= IC). Typically, the IC

analysis is binary in the sense of breaking up a unit into two lower-level units, but a

constituent may also have three (or more) parts, as in A, B, and C. Also discontinuous

constituents are allowed: an indefinite number of adjectives may occur inside the constituent

the ... king. 

 Harris (1946) recommends choosing those descriptive devices “in terms of which we get

the most convenient total description”. As in glossematics, the ultimate criterion for preferring

one description to another is the overall simplicity. 

26

 Echoing Varro, Bloomfield (1933: 275) notes that, as far as speech-forms are concerned, 

“the possibilities of combination are practically infinite”. Hockett (1948) makes the same

point: the linguist must “account also for utterances which are not in his corpus ... he must be

able to predict what other utterances the speakers of the language might produce”. Hockett

(1949) adds that grammatical description constitutes a “recurrent cycle of prediction,

checking, gathering of new data, modification of predictions, and rechecking”. Harris (1951:

17) agrees: “when the linguist offers his results as a system representing the language as a

whole, he is predicting that the elements set up for his corpus will satisfy all other bits of

talking in that language.”  

 Bloomfield (1926) purports to present his language-conception in the axiomatic format of

metalinguistic “assumptions” and “definitions”. In a similar way, Harris (1951: 372–373)

assumes that his grammar-conception amounts to a “deductive system with axiomatically

defined initial elements and with theorems concerning them. The final theorems would

indicate the structure of the utterances of the language”. Such a description would “enable

anyone to synthesize or predict utterances in the language”. In practice, Harris’s “deductive

system” is much too intricate to be applied by just “anyone”. Hockett (1954) uses the term

generate instead of synthesize.  

 Harris (1951) views utterances as purely physical phenomena which should be described

with no recourse to meaning. This position is so extreme that none of the contemporary

reviewers of the book was willing to accept it wholesale (cf. Hymes and Fought 1981: 146–

147). “No one has ever really done linguistics in that way”, as Chafe (1994: 14) was to

observe later.  

 Saussure (1916), Hjelmselv (1943), Sapir (1921), Bloomfield (1933), and Harris (1951) are

not concerned with any of those psychological or biological mechanisms that ‘support’

language. By contrast, Hockett (1948) claims that insofar as the linguist is acting like a

genuine scientist, his “analytical process thus parallels what goes on in the nervous system of

a language learner, particularly, perhaps, that of a child learning his first language” . Thus, it

is the speaker, and not the linguist, who creates linguistic structure: “The child in time comes

to behave the language; the linguist must come to state it”.        

 In the 1940s Harris was perceived as advocating a conception of “linguistics as a game”, to

such an extent that Hockett (1968: 35) could later refer to “Harris’s theoretical nihilism”. But

Harris (1954) clearly involves a change of attitude: “Mathematical and other methods of 

27

arranging data are not a game”. Nor is language viewed anymore as an entirely self-contained 

entity: “the position of the speakers is after all similar to that of the linguist”. 

 In his 1955 dissertation Noam Chomsky (1928–) subscribes to Bloomfield’s antimentalism 

(p. 86) and declares his concern with the “physical properties of utterances” in the sense of

Harris-type distributional analysis (p. 127, p.63, n. 1). Unlike his predecessors, he restricts his

data to English. His 1957 Syntactic Structures is based on a set of 39 self-invented sentences.

Some of these are grammatical and either simple like The man comes or compound like John

enjoyed the book and liked the play. Others are either just ungrammatical like *Lunch is eaten

John or very ungrammatical like **Of admires John. The grammar that he outlines has two

components (apart from morphophonemics). Phrase structure (= PS) rules assign a treestructure

derivation to strings identifiable as simple

sentences,

while transformational

(=

T)

rules

operate either on one string with a specific derivation

or on two such strings to generate

more

complex

sentences. T-rules

are needed because PS-rules (wrongly

identified with IC

analysis)

cannot handle discontinuous

constituents, active–passive

relations, and NP-and-NP

or

S-and-S

conjunctions: “[T-rules] lead to an entirely

new conception of linguistic structure”

(p.

44). The resulting bipartite grammar

is supposed

to give an adequate description to all

grammatical

sentences, whether simple

or complex,

of any language. 

 But this is the real novelty: “the set of ‘sentences’ of some formalized system of

mathematics can be considered a language [and vice versa]” (Chomsky 1957: 13). More

precisely, “the idea of a generative grammar emerged from an analogy with categorial

systems in logic. The idea was to treat grammaticality like theoremhood in logistic systems

and to treat grammatical structure like proof structure in derivations” (Katz 1981: 36,

emphasis added). This is a striking illustration of how scientific discovery is driven by

analogy (see Itkonen 2005: 16–19, 190–196). As a result, a generative grammar turns out to

be, technically speaking, an axiomatic system (cf. Wall 1972: 197–212). The initial S-symbol

qualifies as the sole axiom while the function of inference rules is performed, first, by

structure-expanding PS-rules and, second, by structure-changing T-rules. It is only the

terminal string of the entire derivation that is supposed to qualify as grammatical. (An

axiomatic theory, by contrast, has a property, i.e. either empirical truth or validity, which is

transferred from axioms to theorems by inference rules.) Thus, in Syntactic Structures, it is

the formalization of morphosyntax that is genuinely new. The grammatical analysis itself is

thoroughly traditional, starting with the Aristotelian NP + VP dichotomy.      

28

 What is the methodological status of generative description? This question is answered in 

very dissimilar ways. First, generative description is identified with conceptual analysis as

practiced in philosophy: “We thus face a familiar task of explication of some intuitive concept

— in this case, the concept ‘grammatical in English’” (Chomsky 1957: 13). But then it is the

natural sciences that provide the model: “[A chemical theory] might be said to generate all

physically possible compounds just as a grammar generates all grammatically ‘possible’

utterances” (p. 48). In the same vein, grammatical categories (like NP or T) are equated with

hypothetical constructs (like ‘electron’), and grammatical rules (like NP → T + N and T →

the) are equated with laws of nature (p. 49). 

 Thus, Chomsky does not merely perpetuate the axiomatic tradition in linguistics, but

makes it fully explicit. The commitment to axiomatics is also evident from the central role

that he, like Hjelmslev, ascribes to simplicity: “the only ultimate criterion in evaluation is the

simplicity of the whole system” (p. 56). He even speculates (p. 51) about an algorithm (called

“evaluation procedure”) for choosing the simplest grammar from among several alternatives.

No such algorithm has been or, most probably, will ever been found: “Let us call a theory

which obeys Ockham’s razor ... functionally simple. ... Ockham’s razor seems difficult or

impossible to formalize as an algorithm” (Putnam 1981: 133). 

 After 1957, the generative grammar of English as envisaged by Chomsky has undergone

radical modifications. It has also been given radically new interpretations. In Aspects of a

Theory of Syntax (1965) it became both a mentalistic theory and the basis for a universal

grammar. In the mid-70s it became a biological theory: now language is a module (or

“organ”) of the biologically-based mind, and syntax is the central module of language. Yet

Chomsky’s approch to data has not changed at all. In the 21st century he still investigates

distributional properties of such English sentences as his own intuition deems to be either

grammatical or ungrammatical. 

 For Paul (1880, esp. Ch. III), variation was a necessary part of linguistic change. While

endorsing the non-statistical nature of synchronic grammatical description (cf. above),

Bloomfield (1933) was fully aware that any realistic account of linguistic change would be

statistical in character: “These changes could be observed only by means of genuinely

statistical observation through a considerable length of time; for want of this, we are ignorant

of many matter concerning linguistic change” (p. 38). But the situation has changed

dramatically since Bloomfield’s time. In the mid-1960s sociolinguistics, through the 

29

‘variationist paradigm’, was developed to deal with exactly those types of phenomena that 

Bloomfield placed beyond the ken of linguistics.

 To be sure, what is at issue is not just variation in change, but variation in general. As 

formulated by William Labov (1927–), “the problem is how observation and experiment are

to relate to intuitive data” (1975: 54). He provides a balanced answer with many gradations,

which may be summarized as follows. As far as the so-called clear cases are concerned, “we

can ... study the social aspect of language through the intuitions of one or two individuals” (p.

9). Consider these sentences: That John told him was a shame and *John told him was a

shame. Here the difference in grammaticalness is so clear-cut that “no one has as yet found

any disagreements that would move us to begin a program of observation and experiment” (p.

8). Outside the clear cases, however, linguistic intuition is not enough, and it is here that a

statistical approach becomes a necessity. Two principal cases may be distinguished: either

the data are so complex or otherwise indeterminate that intuitions diverge; or the data are

constituted by frequencies of occurrence, and here intuition is unreliable by definition. Rather

than being ‘free’, variation typically turns out to be conditioned, and thus explained, by

extraneous linguistic or non-linguistic factors. 

 For the most part, linguistic distinctions are not categorical but gradual. This is true of,

e.g., ‘grammatical vs ungrammatical’, ‘obligatory vs optional’, ‘noun vs not-noun’, etc. (But

notice that the difference between the two extremes of a continuum is absolute, not relative.)

Just as categorical distinctions are based on two-valued logic, continuous distinctions are

based on fuzzy logic (

which is in turn a direct descendant of Jan Łukasiewics’s manyvalued

logic;

cf. Kosko 1994)

.

Any adequate linguistic theory

should be able to deal not

just

with variation but also with fuzziness. 

 Such and similar concerns brought it about that the methodological status of linguistics in

general, and of generative linguistics in particular, became the object of intense controversy in

mid- and late 1970s, see the collections of articles such as Cohen (1974), Cohen and Wirth

(1975), and Perry (1980). 

 In the 1970s and 1980s Chomsky claimed that linguistics is part of cognitive psychology

and cognitive psychology is part of biology. Montague (1974) and Katz (1981) countered by

claiming linguistics to be a branch of logic or mathematics. Many felt that generativism

offered a much too impoverished view of what cognition is really about. As a result, the

school of Cognitive Linguistics came into being: “it opened language description to a rich

new landscape of conceptual phenomena and mechanisms interrelated in multiple ways with 

30

the whole of human experience and shaped in accordance with patterns of human 

imagination” (Harder 2007: 1247–1248).

 The ‘first-generation’ cognitivists like

Lakoff (1987), Langacker (1987), and Talmy 

(2000) m

ake use of exactly the same data basis as their generativist predecessors, namely

their own respective intuitions (almost exclusively of English). To members of the ‘secondgeneration’

cognitivists, this

is not enough. Rather, contacts have

to be (re)established with

main-stream

psychology by adopting a set of statistical-cum-experimental

techniques (cf.

Gonzalez-Marquez

et al. 2007). On the other hand,

the exclusive concern with the individual

mind

has to be transcended

by squarely accepting

the

primarily

social nature

of language (cf.

Zlatev

et al. 2008). 

5. Diachronic and Typological Linguistics       

August Schleicher (1821–1868) was a champion of diachronic linguistics: “If we do not know

how something has come into being, we do not understand it.” But he is no less a champion of

typological linguistics: “We do not comprehend the essence of a language unless we relate it

to other languages.” In his 1850 Linguistische Untersuchungen (quoted from Arens 1969:

251–255) he follows the preceding tradition insofar as he divides the object of diachronic

linguistics into two clearly separate parts. The prehistory of languages is characterized by

progress whereas their documented history is characterized by decline. The former contains

an ascending three-stage development ‘monosyllabic structure (Einsilbigkeit) >

Agglutination > Flexion’, on the analogy of ‘mineral > plant > animal’. The latter contains a

degeneration of synthetic into analytic languages, as had already been proclaimed by August

von Schlegel (1767–1845) in 1818 (cf. Arens 1969: 189).   

 In 1850 Schleicher regards diachronic linguistics as a natural science pure and simple, and

in 1863 he reasserts this position. Methodological monism is unmistakeably the order of the

day (“Die Richtung der Neuzeit läuft unverkennbar auf Monismus hinaus”; op. cit., p. 259). 

 In his 1875 book The Life and Growth of Language William Dwight Whitney (1827–1894)

too recognizes what is the Zeitgeist “in these days when the physical sciences are filling

men’s minds with wonder at their achievements” (p. 310–311). But he prefers to go against

the current: “There is no way of claiming a physical character for the study of [linguistic]

phenomena except by a thorough misapprehension of their nature” (p. 311). Why is it a

misapprehension? Because no linguistic change “calls for the admission of any other efficient

force than reasonable action, the action for a definable purpose, of the speakers of language” 

31

(p. 144). And reasonable actions flow from “the faculty of adapting means to ends, of 

apprehending a desirable purpose and attaining it” (p. 145). Every change is the result of a

“choice” prompted by the “human will” even though all this happens “without any reflective

consciousness” (p. 146).    

 In essence, Hermann Paul’s conception of linguistic change is the same as Whitney’s. All

languages share the goal (Ziel) of establishing a one-to-one correspondence between

meanings and forms (“für das funktionell Gleiche auch den gleichen lautlichen Ausdruck zu

schaffen”, p. 227). But this “symmetry of the form system” is constantly being destroyed by

sound change. The principal means (Mittel) of mending the damages caused by sound change

is analogy. Thus, from a bird’s-eye view, the history of any language is an eternal tug-of-war,

governed by a means-ends strategy, between two opposite tendencies, namely material and

intellectual (p. 198). In the same spirit Roman Jakobson (1896–1982) later outlines a “meansend

model

of language” (1990: 56–61).     

 Saussure (1916: 115–116) regarded linguistic change as affecting single units only, which

means that diachronic linguistics, unlike synchronic linguistics, cannot be a systematic or

structure-based discipline. This is clearly not true, as was noted by Roman Jakobson in his

1931 article Prinzipien der historischen Phonologie. The prime example of diachronic

structuralism is given by André Martinet (1908–1999) in his 1955 book Économie des

changements phonétiques, which adds an important correction to Paul’s view of linguistic

change. Instead of merely being a destructive force, sound change contains its own tug-ofwar:

on the one hand, there is the intellectual

effort to maintain

the “stability” of phonological

systems;

on the other, due to “the inertia and the

asymmetry

of the articulatory organs”,

there

is

the tendency to disrupt this stability (p. 89, 101).

Martinet notes (p. 67,

97) that he was to

some

extent anticipated by Trubetzkoy who mentioned

in 1933 that sound change is

characterized

by a “tendency towards harmony”.

    If linguistic change is explained as being a means that speakers (subconsciously)

consider adequate for achieving a given goal, then the corresponding explanation must

qualify as teleological. (Alternative designations are “functional explanation”, “explanation

by reasons”, and “rational explanation”.) This type of explanation may be difficult to accept

because it is nonexistent in the natural sciences. This is why Martinet hesitates (p. 18, 97) to

call his explanations “teleological”, although he otherwise agrees (p. 16–17) with Paul that

what he is trying to do is give causal explanations of linguistic change. There is another

problem as well. Natural-science explanations are not just non-teleological, but they are also 

32

nomological, i.e. based on either deterministic or statistical laws. Thus Lass (1980), for 

instance, sees little or no value in diachronic linguistics because of its non-nomological

character. The other option is, obviously, to accept the basic fact that linguistic behavior (like

intelligent human behavior in general) is prompted by a non-nomological type of causation

(see Itkonen 1983).    

 Since the 1990s it has become fashionable to view linguistic change in Darwinist terms.

This is an interesting return to the position that Schleicher held in the 1860s. But the analogy

between biological evolution and linguistic change is clearly defective.  In general, those who

subscribe to Darwinism are willing to admit that linguistic change involves some sort of

problem solving. In maximally general terms, adopting a given means is a solution to the

problem of attaining the end. But this produces a contradiction: “No evolutionary change of

any kind came about through the application of intelligence and knowledge to the solution of

a problem” (Cohen 1986: 125). 

  Von der Gabelentz and Sapir are among the pioneers of typological thinking. The

foundations of modern typological linguistics were laid by Joseph Greenberg (1915–2001) in

the two books edited by him, Universals of Language (2nd ed. 1966) and the 4-volume

Universals of Human Language (1978). The basic thesis is that, as conjectured by Plato and

Varro, and confirmed by cross-linguistic regularities, grammatical categories and/or

constructions have been shaped, and should accordingly be explained, by the functions they

serve. For instance, such explanatory notions have been proposed as iconicity, economy,

cognitive salience, and animacy hierarchy (i.e. 1/2 SG > human > animate > inanimate). It

seems clear enough that the access to such notions is provided, ideally, by the linguist’s

empathy, or his/her capacity to reconstruct those thought processes that the speakers of the

various languages must have undergone.  

 This view is as old as typological linguistics itself: “we need to put ourselves precisely in

the nomenclator’s place, apprehending just his acquired resources of expression and his habits

of thought and speech as founded on them; realizing just his insight of the new conception

and his impulses to express it” (Whitney 1875: 143). According to Paul (1880: 349), such a

process as suffixation must be made “psychologically understandable” [psychologisch

begreifbar]; this is what historical explanation amounts to in practice. Havers (1931: 211–

212) demands that, however difficult it may be, one must get rid of the thought habits of one’s

mother tongue in order to achieve a “total empathy (Einfühlung) with the mental reality of the

alien language”. Hockett (1955: 147) correctly points out that “we know of no set of 

33

procedures by which ... a machine could analyze a phonologic system. ... The only rules that 

can be described are rules for a human investigator, and depend essentially on his ability to

empathize”. Talmy Givón (1936–) concludes this list of methodologically crucial citations:

“the scientist merely recapitulates the bio-organism” (2005: 204). 

 Whitney (1875) describes the genesis of new word forms as follows: “suffixes of

derivation and inflection are made out of independent words, which ... gradually lose their

independent character, and finally come to be ... mere subordinate elements ... in more

elaborate structures” (p. 124). This process, which of course encompasses more than just

suffixation, is called Komposition (= ‘condensation’) by Paul (1880), who characterizes it as

“the normal way that anything formal emerges in a language” (p. 325). Today it is called

grammaticalization. It was the central topic of the 19th century linguistics, launched by

Franz Bopp. Eclipsed by synchronic structural and/or generative analysis during most of the

20th century, it was rediscovered in the 1970s. 

 The process of grammaticalization has been shown to follow a large set of welldocumented

“paths”, many

of which are listed by

Heine and Kuteva (2002). This means

that

linguistic

change, though still non-nomological,

has become

predictable

at least

to some

extent.

Now that grammaticalization

theory has been

integrated into typological

linguistics, it

is

generally agreed that any synchronic

description not

embedded

in a wider historical context

is

just as deficient

as any description without

typological background. “The

linguist who

asks

‘Why?’

must

be a historian” (Haspelmath

1999: 205).

Following the spiral (rather than cycle)

of

history, we have returned — ‘at a higher

plane’ — to Schleicher’s and Paul’s

position.              

6. Conclusion 

In what precedes, the need for, and the nature of, explanation as well as the idea of axiomatics

were shown to be central themes in recounting the philosophy of linguistics. Katz (1981: 52,

64–68) outlines the notion of an “optimal grammar” which, free from any non-grammatical

(i.e. psychological, sociological or neurological) constraints, should be based on the notion of

overall simplicity. Here ‘optimal’ clearly equals ‘axiomatic’, and Pāṇini’s grammar, “the most

complete generative grammar of any language yet written” (Kiparsky 1993: 2912), is so far

the most likely candidate for being an optimal grammar. A few qualifications, however, need

to be added. 

 Even within formal logic, axiomatics is by no means the only alternative. For instance, the

so-called dialogical logic dispenses with the “traditional recourse to the axiomatic method” 

34

(Lorenzen and Lorenz 1978: 19). Being a “reconstruction” of the norms of everyday language 

and thus embodying the social nature not just of language but also of logic, it is clearly

superior to axiomatic logic (see Itkonen 2003: Ch. IV). 

 Within linguistics, furthermore, the axiomatic ideal applies only to the grammatical theory,

i.e. the subdiscipline discussed in Itkonen (1978). Everywhere else, that is, in diachronic

linguistics, sociolinguistics, and psycholinguistics, causal explanation is to be preferred. This

view was adumbrated by Hermann Paul, and it also agrees with the more recent developments

within the philosophy of the natural sciences: “A scientific theory admits of many different

axiomatizations, and the postulates chosen in a particular one need not, therefore, correspond

to what in some more substantial sense might count as the basic assumptions of the theory, ...”

(Hempel 1970: 152). This “more substantial sense” refers to the causal structure of the

research object, to be described by a corresponding causal model.        

 Salmon (1984) comes to the same conclusion: “It now seems to me that the statistical

relationships specified in the S[tatistical]–R[elevance] model constitute the statistical basis

for a bona fide scientific explanation, but that this basis must be supplemented by certain

causal factors in order to constitute a satisfactory scientific explanation” (p. 34, original

emphasis). It is this aspect of linguistics that is focussed on in Itkonen (1983).   

 

References 

Allan, Keith. 2010. The Western classical tradition in linguistics. 2nd edn. London and

Oakville: Equinox. [First edn 2007] 

Apollonius Dyscolus. 1981.The syntax of Apollonius Dyscolus. Translated, and with a

commentary by Fred W. Householder. Amsterdam: John Benjamins. 

Arens, Hans. 1969. Sprachwissenschaft: Der Gang ihrer Entwicklung von der Antike bis zur

Gegenwart, Band 1. Frankfurt am Main: Athenäum. [First edn 1955] 

Arens, Hans. 1984. Aristotle’s theory of language and its tradition. Amsterdam: John

Benjamins. 

Aristotle. 1941. The basic works of Aristotle. Edited and with an introduction by Richard

McKeon. New York: Random House. 

Arnauld, Antoine and Claude Lancelot. 1968. Grammaire générale et raisonnée. Gen ve:

Slatkin Reprints. [First published 1660] 

35

Arnauld, Antoine and Pierre Nicole. 1965. La logique ou l’art de penser. Éd. critique par P. 

Clair et F. Girbal. Paris: Presses universitaires de France. [First published 1662, final

version 1683] 

Auroux, Sylvain. 1989. Introduction. In Histoire des idées linguistiques, Tome 1: La

naissance des métalangues en occident et en orient. Li ge-Bruxelles: Mardaga. Pp. 13–37. 

Becker, Karl Ferdinand. 1841. Organism der Sprache. 2nd ed. Frankfurt am Main: Verlag

von G.F. Kettembeil. 

Bekkum, Wout van, Jan Houben, Ineke Sluiter and Kees Versteegh. 1997. The emergence of

semantics in four linguistic traditions: Hebrew, Sanskrit, Greek, Arabic.

Amsterdam/Philadelphia: John Benjamins. 

Blackburn, Simon. 1984. Spreading the word. Oxford: Oxford University Press.

Bloomfield, Leonard. 1926. A set of postulates for the science of language. Language 2: 153–

64.

Bloomfield, Leonard. 1933. Language. New York: Holt, Rinehart and Winston.

Boas, Franz. 1911. Introduction. In Handbook of American Indian languages, Bureau of 

American Ethnology, Bulletin 40, Part 1. Washington, D.C.: Smithsonian Institution. Pp.

1–83.  

Boethius de Dacia. 1980. Godfrey of Fontaine’s abridgement of Boethius de Dacia’s

MODI

SIGNIFICANDI SIVE QUAESTIONES SUPER PRISCIANUM MAIOREM

. Edited, translated, and

introduced by Charlene McDermott. Amsterdam: John Benjamins.  

Chafe, William. 1994. Discourse, consciousness, and time. Chicago: Chicago University

Press. 

Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton.

Chomsky. Noam. 1965. Aspects of a theory of syntax. Cambridge, MA: The MIT Press.

Cohen, Daniel (ed). 1974. Explaining linguistic phenomena. Washington–London: 

Hemisphere Publishing Corporation.

Cohen, Daniel and Jessica Wirth (eds). 1975. Testing linguistic hypotheses. Washington–

London: Hemisphere Publishing Corporation.

Cohen, L. Jonathan. 1986. The dialogue of reason: An analysis of analytical philosophy. 

Oxford: Clarendon Press. 

36

Elffers-van Ketel, Els. 1991. The historiography of grammatical concepts. Amsterdam – 

Atlanta, GA: Rodopi.

Evans, Jonathan St. B.T. 1982. The psychology of deductive reasoning. London: Routledge 

and Kegan Paul. 

Fung Yu-lan. 1952. A history of Chinese philosophy I. Transl. by Derk Bodde. Princeton: 

Princeton University Press.

Gabelentz, Georg von der. 1901. Die Sprachwissenschaft: ihre Aufgaben, Methoden und 

bisherigen Ergebnisse. Leipzig: T.O. Weigel Nachfolger. [First edn 1891] 

Givón, T. 2005. Context as other minds: The pragmatics of sociality, cognition and 

communication. Amsterdam/Philadelphia: John Benjamins. 

Gonzalez-Marquez, Monica, Irene Mittelberg, Seana Coulson and Michael J. Spivey (eds). 

2007. Methods in cognitive linguistics. Amsterdam/Philadelphia: John Benjamins.

Goody, John. 1977. The domestication of the savage mind. Cambridge: Cambridge University 

Press.

Greenberg, Joseph H. (ed). 1966. Universals of language, 2nd edn. Cambridge, Mass.: The 

MIT Press. [First edn 1963]

Greeberg, Joseph H. (ed). 1978. Universals of human language. Volumes 1–4. Stanford: 

Stanford University Press.  

Harder, Peter. 2007. Cognitive linguistics and philosophy. In The Oxford handbook of 

cognitive linguistics. Ed. by Dirk Geeraerts and Hubert Cuyckens. Oxford: Oxford

University Press. Pp. 1241–1265. 

Harris, Zellig S. 1946. From morpheme to utterance. Language 22: 161–83.

Harris, Zellig S. 1951. Methods in structural linguistics. Chicago: University of Chicago 

Press.

Harris, Zellig S. 1954. Distributional structure. Word 10: 146–62.

Haspelmath, Martin. 1999. Optimality and diachronic adaptation. Zeitschrift für 

Sprachwissenschaft 18/2: 180–205.

Havers, Wilhelm. 1931. Handbuch der erklärenden Syntax. Heidelberg: Carl Winters 

Universitätsbuchhandlung. 

37

Heine, Bernd and Tania Kuteva. 2002. World lexicon of grammaticalization. Cambridge: 

Cambridge University Press.

Hempel, Carl G. 1970. On the ‘standard conception’ of scientific theories. In Minnesota 

studies in the philosophy of science IV, ed. by M. Radner and S. Winokur. Minneapolis:

University of Minnesota Press. Pp. 145–160. 

Hjelmslev, Louis. 1963. Prolegomena to a theory of language. 2nd ed. Transl. by Francis J.

Whitfield. Madison: University of Wisconsin Press. [First published in 1943] 

Hockett, Charles F. 1948. A note on structure. International Journal of American Linguistics

14: 269–71. 

Hockett, Charles F. 1949. Two fundamental problems in phonemics. Studies in Linguistics 7:

29–51. 

Hockett, Charles F. 1954. Two models of grammatical description. Word 10: 210–231.

Hockett, Charles F. 1955. A manual of phonology. Indiana University Publications in  

Anthropology and Linguistics, Memoir 11.

Hockett, Charles F. 1968. The state of the art. The Hague: Mouton.

Householder, Fred W. 1981. Commentary. In The syntax of Apollonius Dyscolus. Amsterdam: 

John Benjamins.

Husserl, Edmund. 1913. Logische Untersuchungen I–II. 2nd ed. Halle: Niemeyer. [First edn 

1901–1902]

Hymes, Dell and John Fought. 1981. American structuralism. The Hague: Mouton Publishers. 

[First published in 1975]

Itkonen, Esa. 1978. Grammatical theory and metascience. Amsterdam: John Benjamins.

Itkonen, Esa. 1983. Causality in linguistic theory. London: Croom Helm.

Itkonen, Esa. 1991. Universal history of linguistics: India, China, Arabia, Europe. 

Amsterdam/Philadelphia: John Benjamins.

Itkonen, Esa. 2003. Methods of formalization inside and outside both autonomous and non-

autonomous linguistics. University of Turku: Publications in General Linguistics 6.

Itkonen, Esa. 2005. Analogy as structure and process: Approaches in linguistics, cognitive 

psychology and philosophy of science. Amsterdam/Philadelphia: John Benjamins. 

38

Jakobson, Roman. 1931. Prinzipien der historischen Phonologie. Travaux de Circle 

Linguistique de Prague 4: 247–67.

Jakobson, Roman. 1990. Efforts towards a means–end model of language in interwar 

continental linguistics. In Roman Jakobson On language, ed. by Linda R. Waugh and

Monique Bonville-Burston. Cambridge MA: Harvard University Press. Pp. 56–61. [First

published in 1958]  

Johnson-Laird, P.N. 1983. Mental models. Cambridge: Cambridge University Press.

Katz, Jerrold J. 1981. Language and other abstract objects. Oxford: Blackwell.

Katz, Jerrold J. 1985. Introduction. In The philosophy of linguistics, ed. by Jerrold J. Katz. 

Oxford: Oxford University Press. Pp. 1–16.

Kiparsky, Paul. 1993. Pāṇinian linguistics. In The encyclopedia of language and linguistics, 

ed. by R.E. Asher. Oxford: Pergamon Press. Pp. 2918–23. 

Kneale, William and Martha Kneale. 1962. The development of logic. Oxford: Clarendon 

Press. 

Kosko, Bart. 1994. Fuzzy thinking: The new science of fuzzy logic. London:

HarperCollins 

Labov, William. 1975. What is a linguistic fact? Lisse: The Peter de Ridder Press. 

Lakoff, George. 1987. Women, fire, and dangerous things. Chicago: The University of

Chicago Press. 

Langacker, Ronald W. 1987. Foundations of cognitive grammar, Vol I: Theoretical

prerequisites. Stanford CA: Stanford University Press. 

Lass, Roger. 1980. On explaining language change. Cambridge: Cambridge University Press.

Lorenzen, Paul and Kuno Lorenz. 1978. Dialogische Logik. Darmstadt: Wissenschaftliche 

Buchgesellschaft.

Loux, Michael J. 1974. Ockham on generality. In Ockham’s theory of terms: Part I of the 

Summa Logicae. Transl. by Michael J. Loux. Notre Dame: University Press. Pp. 23–46. 

Łukasiewics, Jan. 1935. Zur Geschichte der Aussagenlogik. Erkenntnis V: 111–31.  

Martinet, André. 1955. Économie des changements phonétiques. Bern: A. Francke.

Marty, Anton. 1908. Untersuchungen zur Grundlegung der allgemeinen Grammatik und 

Sprachphilosophie. Halle: Niemeyer. 

39

Masica, Colin P. 1991. The Indo-Aryan languages. Cambridge: Cambridge University Press.

Montague, Richard. 1974. Formal philosophy. New Haven: Yale University Press.

Ockham, William. 1974. Ockham’s theory of terms: Part I of the Summa Logicae. Transl. by 

Michael J. Loux. Notre Dame: University Press.

Ockham, William. 1980. Ockham’s theory of propositions: Part II of the Summa Logicae. 

Transl. by Alfred J. Freddoso and Henry Schuurman. Notre Dame: University Press. 

Owens, Jonathan. 1988. The foundations of grammar: An introduction to medieval Arabic 

grammatical theory. Amsterdam/Philadelphia: John Benjamins.

Pap, Arthur. 1958. Semantics and necessary truth: An inquiry into the foundations of analytic 

philosophy. New Haven: Yale University Press.

Parret, Herman. 1976. Le débat de la psychologie et de la logique concernant le langage: 

Marty et Husserl. In History of linguistic thought and contemporary linguistics, ed. by

Herman Parret. Berlin/New York: Walter de Gruyter. Pp. 732–71. 

Paul, Herman. 1975. Prinzipien der Sprachgeschichte. 5th edn. Tübingen: Max Niemeyer

Verlag. [First edn 1880] 

Paul, Hermann and Walther Mitzka. 1960. Mittelhochdeutsche Grammatik. Tübingen: Max

Niemeyer Verlag. [First edn 1881] 

Perry, Thomas A. (ed). 1980. Evidence and argumentation in linguistics. Berlin/New York:

Walter de Gruyter.  

Plato. 1963. The collected dialogues, ed. by Edith Hamilton and Huntington Cairns.

Princeton: Princeton University Press. 

Putnam, Hilary. 1981. Reason, truth and history. Cambridge: Cambridge University Press.

Reichenbach, Hans. 1947. Elements of symbolic logic. New York: The Free Press.

Rubin, David C. 1995. Memory in oral traditions. Cambridge: Cambridge University Press.

Salmon, Wesley C. 1984. Scientific explanation and the causal structure of the world.  

Princeton, N.J.: Princeton University Press.

Sapir, Edward.  1921. Language: An introduction to the study of speech. New York: Harcourt, 

Brace. 

40

Sapir, Edward. 1949. The psychological reality of phonemes. In Selected writings of Edward 

Sapir in language, culture and personality, ed. by David G. Mandelbaum. Berkeley:

University of California Press. Pp. 46–60. [First published 1933] 

Saussure, Ferdinand de. 1962. Cours de linguistique générale. Paris: Payot. [First published

1916] 

Sigerus de Cortraco. 1977. Summa modorum significandi. Sophismata. New edition, with an

introduction by Jan Pinborg. Amsterdam: John Benjamins. 

Spang-Hanssen, Henning. 1966. On the simplicity of descriptions. In Readings in linguistics

II, ed. by Eric P. Hamp, Fred W. Householder and Robert Austerlitz. Chicago/London: The

University of Chicago Press. Pp. 234–41. [First published 1948] 

Talmy, Leonard. 2000. Towards a cognitive semantics, 2 vols. Cambridge MA: MIT

Press.

 

Trubetzkoy, N.S. 1958. Grundzüge der Phonologie. Göttingen: Vandenhoeck and Ruprecht.

[First published in 1939] 

Varro, Marcus Terentius. 1938. De lingua latina I–II. Varro on the Latin language. Transl. by

Roland G. Kent. London: Heinemann. 

Wall, Robert. 1972. Introduction to mathematical linguistics. Englewood Cliffs: PrenticeHall.

 

Wedgwood, Ralph. 2007. The nature of normativity. Oxford: Oxford University Press.

Wells, Rulon S. 1947. Immediate constituents. Language 23: 81–117. 

Zlatev, Jordan, Timothy P. Racine, Chris Sinha and Esa Itkonen (eds). 2008. The shared 

mind: Perspectives on intersubjectivity. Amsterdam/Philadelphia: John Benjamins.