Table Of Contents

1. Introduction

2. Speech & Writing

3. Sources of the Written Language Bias in Linguistics

4. On the Scope of our Claims

5. Linguistics & the Overall Theory of Language

6. Grammar

7. Semantics

8. Phonology

9. Linguistic Communication

10.Language Acquisition

11 Linguistic Variation

12 Epilogue




'There is, incidentally, nothing particularly surprising about the fact that conventional orthography is, as these examples suggest, a near optimal system for the lexical representation of English words. The fundamental principle of orthography is that phonetic variation is not indicated where it is predictable by general rule' (Chomsky & Halle 1968:49)


8.1 Phonological structure

Phonetic behavior and the resulting acoustic signals are continuous dynamic phenomena. The various phonetic gestures involved in speech production overlap and have no abrupt onsets and offsets. Yet it is generally assumed that phonological structures (phonological forms, phonological representations) underlying speech consist of linear sequences of discrete, static segments. This implies that phonological structures would be structurally similar to strings of alphabetic letters. Such letters are in fact used as phonological notation.

The use of alphabetic writing as the metalanguage of phonology is something which may be assumed to have a significant impact on our theories of phonological structure. Before going on to a discussion of this matter, we should notice, however, that not only the abstract, underlying phonological structure of words is notated by means of discrete graphic symbols. We also use more "concrete" representations of the pronunciations of words and utterances, so-called (narrow) phonetic representations, which are likewise couched in a (modified) letter notation. This means, in all probability, that our view regarding phonetic structure is also influenced by the outer form of this written metalanguage. In any case, it would be utterly naive to believe that phonetic transcriptions, no matter how "narrow" they are, are some kind of mechanically computable, "objective" representations (or reflections) of the phonetic signals. On the contrary, they are the result of a conventional transformation of speech into writing and we need have access to implicit (conventional) rules in order to be able to convert them "back" into speech. It seems probable, says Ladefoged (1967:52), "that our lack of knowledge of what we are doing when we make phonetic transcriptions is actually hampering our own work as descriptive linguists".

The phonemic principle and the idea of the double articulation of spoken language are probably historically dependent on the existence of alphabetic writing. Thus, if the continuous and varying stream of behavior has to be notated in writing, there arises a need for an economic set of discrete signs, e.g. letters or other symbols (such as pictures). Therefore, a practical notational system presupposes an analysis in terms of segments of some sort. The next step in the argumentation implies that these segmental units are not only workable units of analysis, they are in fact inherent properties of the subject matter; hence phonologists discovered that there were in fact segments underlying overt behavior.

It would be stupid to deny that the idea of underlying segments has some kind of basis in speech production and perception. First of all, the drive towards categorization applies to the perception of speech as well. For this and other reasons patterns and routines are developed also in speech production, the same motor elements tend to be used in the articulation of all words in the language. There insubstantial evidence for units like syllables, syllabic constituents (onsets, nuclei, offsets), vowels and consonants as units of production; common slips of the tongue (such as) 'spictly streaking' for 'strictly speaking', 'strunction and fucture' for 'structure and function', 'lawn drawn' for 'line drawn' are but one type of evidence. Nevertheless, it is no doubt true that writing and the ability to read and write enhance our experience of speech as being composed of segments. What is at stake here is not the general idea that vowels and consonants are components of speech, but rather the much stronger hypothesis inherent in most phonological theories, i.e. that the phonological structure of a word is just a linear sequence of non-overlapping segments. Fowler formulates the basic point of this "strong segment theory" like this:

Segments in a planned sequence are discrete in the sense that (abstractly stated), their boundaries are straight lines perpendicular to the time axis, so that the terminus of one segment is the beginning of the next segment". (Fowler 1980:116)

According to mainstream phonological theory, each segment is a bundle of simultaneous features. Such a segment sequence is a basically spatial (rather than temporal) organization of thing-like phonological units arranged in a before-after sequence analogous to the left-right sequences of conventional orthography and conventional phonetic notation. In an extreme version, this theory excludes the possibility that suprasegmental features and syllable structure are phonologically significant.

If the phonologist's view of phonetic structure is influenced by the perspective formed by alphabetic writing, this is true of the layman's thinking about speech to an even greater extent. Aside from the fact that sounds (phonemes) and letters are hopelessly mixed up in the linguistic thinking of most laymen, it is clear that writing distorts our phonetic intuition and make us deaf to certain phonetic realities, notably those which have no counterpart in common orthography. For example, phonetically untrained 1isteners do not hear the difference in aspiration between the p:s of 'peak' vs. 'speak', nor do they hear the vowel murmur after the stop in a stop-liquid initial cluster in e.g. 'please' and prayed (which makes them almost homophonous in certain speaking styles with 'police' and 'parade'). Furthermore, the relative inability to perceive prosodic phenomena is another case in point. One may also mention the wide-spread belief that words are regularly separated by silent intervals (pauses) in speech, in analogy with the empty spaces of written texts; actually such pauses are relatively infrequent, which can be seen at any registration of the acoustic signal.

Literacy may conceivably have effects on speakers implicit understanding of other aspects of phonology too, e.g. morphophonology. 2) There is some evidence from various psycholinguistic experiments (recall of nonsense forms, cf. Myerson 1975) that subjects who can read and write English well and thus are sufficiently acquainted with English orthography tend to associate pairwise such heterogeneous vowels as English

rai3- ~J,

ti~ ~ ~ ta~ etc.

Young children, on the other hand, seem to base similarity judgments on phonetic properties (as demonstrated in spontaneously invented spellings, Read 1971). To the extent that these results hold true in general, the reason for adult speakers' associations is of course the fact that single vowel letters are used in English orthography to designate pairwise quite different vowels, i.e.

 i for ~1 and ~], a for ~e~l~ and rae3 etc.

 Let us now return to our main theme, the contention that the idea a of double articulation derives some of its motivation from alphabetic writing. It seems to me that this hypothesis receives a certain amount of confirmation from the recent history of the theory of the sign languages of the deaf. Earlier, many people in the field seem to have argued that signed languages are not doubly articulated (Healy 1973). However, later on some scholars (e.g. Stokoe 1960, 1972) "discovered" that they did in fact display double articulation. This discovery seems to have been connected with the invention of discrete written symbols to represent the movement components of gestures (cheremes). When the medium of representation required an analysis in terms of such distinctive features, the theory that these features do in fact exist as inherent parts of the subject matter was immediately naturally at hand.

Some of the insights of phonological theory, e.g. as regards the capacity for meaning differentiation inherent in different sound gestalts, has been implicit in linguistic thinking for centuries, even millennia. In fact, the development of alphabetic writing presupposed some kind of phonological analysis of speech. Also, the practical goals, for which phonology has proved necessary, are largely connected with the need for devising alphabets for new languages. This shows up clearly in the practice of American descriptivists following Boas, Sapir and Bloomfield; Pike gave his book 'Phonemics' (1947) the subtitle 'A technique for reducing languages to writing'. It may well be that even contemporary phonological theory is more suitable for investigating optimal orthographies than for discovering the structure of spoken language.


8.2 L'image acoustique

So far I have identified only one of the ways in which the written language bias has entered phonology, i.e. through alphabetic notations used for phonetic transcription. However, there is at least one more avenue by which the written medium has influenced phonological thinking, viz. through the theories of perception.

Given that phonological structure 1s ascribed some kind of psychological validity, it is reasonable to argue that the phonological structure of words is postulated or constructed by language users on the basis of the sensory information available about acoustic signals. The phonological form of, say, a word is seen as a representation of the phonetic signal, as a structure which is perceptually determined. Instead of representation, other terms have been used, e.g. sound image, mental imaginary counterpart of a sound (Lautvorstellung), or in Saussure's words image acoustique.

Behind these notions lurks an old theory of speech perception. This theory was in its turn modeled on traditional theories of visual perception, according to which we form mental images, inner pictures, of what we see around us. In short, the theory implies that we do not see the outer reality "directly" but we see a picture of it. One of the arguments for such a theory has been that we can see the same stimulus configuration in different ways: we form, as it were, different pictures of it. (After all, the eye does form retinal images). Such theories are fraught with difficulties; among other things, they lead to infinite regressions (who is it that looks at the inner image?). In today's thinking about perception they are being abandoned in favor of theories of "direct" perception, according to which the perceiving agent actively searches for and reinterprets patterns into stimulus configurations (Neisser 1976).

What is important however, is that the theories of visual perception involving mental images were clearly inspired by the existence of real images and pictures, i.e. static thing-like representations on stone, wood, leather or paper. Such pictures may be more or less close to what they represent; at any rate, they may depict the same fragment of reality in different ways. It therefore seemed quite natural to explain our ability to perceive reality in different ways, under different aspects, by assuming that we form different mental images of it. This type of theory was later transferred to the area of auditive perception as well. The same phonetic behavior may be perceived in different ways, e.g. by speakers of different languages, and this could be explained by assuming the existence of different ways of forming mental images, i.e. phonological representations, of the phonetic behavior involved. Again, we see how our picture (sic') of reality is dependent on our experience of written texts and pictorial representations.

It might be added here that a careful analysis of the concept of phonological form shows that an interpretation in terms of sound images is unacceptable on several grounds. This in no way amounts to denying that our knowledge of the phonological structure of speech is perceptual in origin. However, we need not end up with a theory that implies that we hear percepts, or phonological forms, rather than the actual phonetic signals themselves: we do hear phonetic events, but we hear something particular, under certain aspects. The most important aspect implies that the phonetic signals are heard as instances of meaningful communicative behavior in a certain natural language (Linell 1982a).


VIII.3 Phonological rules

It took linguists a very long time to admit unequivocally that letters and sounds (phonemes) were two quite distinct things. However, one may still ask the question whether the ties between the two have been definitely cut off or not. Modern phonology thinks of phonological representations as referring to mental things, i.e. mental counterparts of sounds, auditive images or the like, which are assumed to exist "prior to" the phonetic behavior (VIII.2). Phonological rules, which, by the way, are often assumed to be isomorphic to putative performance processes (VIII.4), are depicted as substitution processes applied to these mental objects. In the allophonic rule of (1), one thing

 ([~]) {nm ~ ~/ f is substituted for another (/m/ or /n/), just as one symbol on paper (<m>) is replaced by another (< ~> ). (cf. Bailey 1979).

(Such a formulation in terms of discrete replacements is particularly inadequate as regards "concrete" rules of hypo- and hyperarticulation, where the variations in pronunciation are gradual rather than absolute (Linell 1979a:188ff)). In general, one may say that phonologists abstract totally from the time-distributed nature of the phonetic signal and perform their analysis of phonological structure as if they were indeed concerned with nothing but strings of graphic symbols. Once the phonetic data have been transcribed, they have taken on all the characteristics of the metalinguistic notation; they have been transformed in the minds of the linguists.

One cannot help noting that the concept of phonological rule must be given a rather odd interpretation in (orthodox) generative phonology (e.g. Chomsky & Halle 1968, Postal 1968). According to traditional phonology, the most reasonable interpretation of the notion "phonological (or phonetic) rule" would be something like "rule for the language-specific exploitation of certain phonetic mechanisms"; thus, a phonological rule covers a regularity in the phonetic behavior pertaining to a specific language or dialect. Such rules are related to habits of behavior, which are partly subject to implicit norms of behavior. These rules are socially given and socially shared, which means that they are also socially acquired; the individual speaker learns the rules by inspecting and analyzing other people's behavior. However, in modern generative phonology we are faced with something entirely different. There, phonological rules are part of large systems of opaque rules operating on vary abstract underlying forms according to extrinsic ordering restrictions (see Chomsky & Halle 1968, and for critical discussion Linell 1979a). Although the outputs of the whole derivations of such a system may be seen as conditions or rules for phonetic behavior, neither the underlying forms nor the specific generative rules as such may be construed as norms of phonetic behavior. While linguistic rules in the traditional sense must be learnable (cf. above), it seems inconceivable that a generative phonological rule system could be acquired by social learning. This point has been raised as a fundamental critique of the whole theory (e.g. by Derwing 1973), but Chomsky draws an entirely different conclusion; he assumes the whole system to be partly derived from very specific innate structures of the human mind (or brain). Furthermore, rules are interpreted as mental (and presumably causally efficient) processes, which again is something completely different from norms of behavior. (This view is implicit in most of generative psycholinguistics, but it is most clearly spelled out in Ratz 1964).

However, upon closer consideration it turns out that generative-phonological rules may in fact be assigned an interpretation which is by and large "rules or norms for behavior". But then we are no longer concerned with the original object of study, i.e., phonetic behavior in speech communication. Instead, we are dealing with meta-rules in the sense of "norms for the linguist's analysis". Note that Chomsky derived his notion of rule from mathematics; a mathematical rule essentially states the conditions under which a given string of symbols may be rewritten as another string of symbols. This is also true of generative-linguistic rules. In fact, a phonological rule amounts precisely to a norm for the phonologist's behavior as he is practicing linguistic analysis; thus, rule (1) says that under certain conditions strings like /mf/ or /nf/ may be substituted for, or rewritten (by the phonologist), as /~f/, or the other way around, under certain conditions /m~f/ may be analyzed (by the phonologist) as one of the underlying strings /mf/ or /nf/. Here we see how the whole theory 18 dependent on the written metalanguage; generative phonological rules state norms for how the (orthodox) phonologist may translate a given string of written symbols into another such string within the framework of a rewriting system that generates phonetic transcriptions, which in turn are devices by which the structure of phonetic behavior has been converted into writing.


8.4 The relation between phonology and Phonetics

Current theories of sound structure assume that there are two different, purportedly real modes of existence of speech sounds, i.e. the phonological strings of discrete, static and context-free segments, and the continuous, dynamic, phonetic behavior. (Note, however, that also the latter phonetic behavior is conventionally transcribed in discrete letter notation (VIII.1)). This necessarily generates a number of problems for the theories of speech production and perception; How should these two seemingly contrary conceptions be reconciled? What role do phonological forms play in speech performance? These issues have been extensively dealt with elsewhere, e.g. in Linell (1982a), and I will therefore confine myself to a few brief statements here.

According to the standard view the phonetic processes of speech production are essentially coarticulatory. This means, in short, that strings of discrete, static segments (conceived of as some kind of mentally real things) are coarticulated in such a way that a dynamically varying continuum arises (e.g. Daniloff & Hammarberg 1973, Hammarberg 1976, Rent & Minifie 1977). The phonetic manifestation process is characterized as "the assignment of phonetic effects to phonological causes" (Hammarberg:356), or as a "conversion" or "translation" of phonological features to "articulatory transitions" (Kent & Minifie:131) ). As Fowler et al (1980) have pointed out, this means that speech production implies a destruction of the clear-cut phonological structure, which turns into "an impoverished acoustic signal n . Conversely, the process of speech perception involves the reconstruction of the phonological form on the basis of a purportedly "defective" signal. It seems to me that well-known values are reflected in such statements. The spoken signal is defective and messy, while the underlying phonological structure is as neat as a string of graphic symbols!

This type of theory with its conception of the relationship between "mental" phonology and "physiological-physical" phonetics creates a number of pseudo-problems, and it refuses to take phonetic behavior at face value, as something which is in fact orderly and structured, although not structured in precisely the way that the "strong segment theory" (cf. VIII.1) predicts. There appears to be evidence, from speech errors and elsewhere, that vowels and consonants, syllables and syllabic constituents, words and parts of words (stems, affixes) are, or may be, phonetic units. Furthermore, we can observe various prosodic phenomena having a suprasegmental and holistic character (cf. V.4). These properties are largely language-specific, and may therefore aptly be thought of as n the phonological structure" of the language involved. But this does not force us to imagine phonological units as some kind of "mental things n which get converted into behavior. That would entail a very problematic "translation theory n of speech communication (IX.l).

Rather, we should prefer another metaphor. Phonetic behavior is a stream of continuous vocal behavior, and the phonological structure may be seen as conditions on, or modulations, on a carrier wave of vocalizations (see Linell 1982a). The brain (or, if you will, the mind) is constantly active, and the perception of a in stimulus pattern as a particular linguistic utterance (with specific phonological, grammatical and semantic properties) would involve inducing a certain change (increase and modification) of the patterns of activity. Such modulations would presumably result from a complex interplay between input stimuli and internally generated constraints (expectations, structural guesses). In the latter phenomenon we can identify the role of linguistic structure:

"In the processing of speech, the function of lexical, syntactic, and phonological structures is that of constraints: they reduce the number of potential guesses available for the identification of the acoustic speech input." (Sajavaara 1980:9)