Navigation – Plan du site

AccueilNuméros27Corpus Linguistics: An Alternative

Texte intégral

1. Philology: linguistics as part of the human sciences

1The 20th century saw the rise of linguistics as a science, an academic discipline comparable to that of physics or chemistry. This new perspective was to a large extent the achievement of Ferdinand de Saussure, the Swiss linguist, who replaced the paradigm of philology, prevalent all over the 18th and the 19th century, but seen as part of the Geisteswissenschaften, of the sciences humaines. The philologists were concerned with texts, mostly texts of ‘dead’ languages. They were primarily interested in the history of languages. History, though, was regarded as the very heart of the sciences humaines. History (re-)constructs the past by interprets the by interpreting past discourses. It is not a hard science. If it were, we could use it to predict the future.

2 Jean-Jacques Rousseau had been one of the philosophes of the Enlightenment movement interested in the origin of language (Essai sur l’origine du langues, published posthumously 1871); another one was Johann Gottlieb Herder (Abhandlung über den Ursprung der Sprache, 1770). Obvious similarities between certain ideas gave rise to the idea that there must be a common origin, and that certain, if not all languages must be related.

3The philologists began to compare languages, not their abstract language systems but their texts and the elements, the words, they found in these texts. They were empiricists; they insisted on working with real language data. They were historically-minded. If two or more languages shared similar words, was there a common origin? How far was it possible to go back? Many of the languages seemed to be related to each other by virtue of their vocabularies: the famous Indo-European language family including, among many other tongues, Sanskrit, Hittite, the Baltic, the Celtic, the Germanic and the Slavic languages, Latin and Greek, but not Hebrew, not the Turk languages, not Hungarian and certainly not Basque.

4Their analyses showed many regularities in the ways words were related across languages in their phonetic (or written) form. Thus it was no surprise that many philologists became fascinated by the possibility of reconstructing former stages of Indo-European or other languages in spite of the absence of data. The ‘laws’ they were trying to discover gave rise to the ideas that languages could be viewed as systems allowing us to ‘predict’ language change. They covered, for instance, sound shifts between more or less stable phases, such as between Old High German and Middle High German, the so-called second sound shift. Their basis was, first of all, the comparison between different dialects or cognate languages that survived into (often) later written representations. But they had problems to cope also with the less than straightforward relationship between the spoken and the written form, problems that could never be satisfactorily resolved. Also it was not so clear how dependable these presumed regularities were. Of course, any evidence of how these words were actually pronounced had to be flimsy. While I went to school, I had to change the pronunciation of Latin c before ae (Caesar), e (cena) or i (Cicero) from [ts] to [k], due to a change of opinion among experts. But the idea that a language, or, at least, its phonetic representation over time, could be studied as something systematic, as something that exists, in some ideal form, outside of texts, outside of real language data, was born. It did not take long before philologists even began to ‘normalise’ the texts they found, for instance the Middle High German epics existing in many variations in more or less complete manuscripts, making them conformant to the ideal language system they had devised. They also reconstructed hypothetical earlier states of Indo-European languages for which there were no written records, and they even wrote texts in them.

5There may be regularities in phonetic change. However, once we look at meaning, we find little regularity. Words on their own are generally fuzzy and ambiguous. This has been shown by many experiments. All attempts to systematically equate the word senses we find in a dictionary with the occurrences of these words in real texts are doomed. This is not so surprising. What lexicographers have to record are generalisations that abstract from the specific meaning a word acquires in a unique context. Most philologists, though, were firmly committed to empirical facts. They wanted their accounts of the registers of Latin, Greek, Hebrew and the sundry of other languages to be reliable for practically every occurrence of a given text in the corpus of the language in question. Thus their glossaries grew to be far more than just dictionaries; they were indeed not only repositories of words but, by embedding them into their specific contexts, interpretations of the of distant cultures, if we take culture as what has been described, by Edward Tyler, as “that complex whole which includes knowledge, belief, art, morals, law, custom and any other capabilities and habits acquired by man as a member of society” (Edward B. Tyler, (1871), Primitive Culture. London: J. Murray, 1). All these facets of culture, indeed all material and textual artefacts, exist, or are made meaningful, in the discourse, and nowhere else. This is what we can learn from the philologists, as we have been forcefully reminded by Edward W. Said in his seminal book Orientalism (London: Penguin 1981, 123-148). As I see it, the philologists were the true forerunners of corpus linguistics.

6Language studies at academic institutions had a hard time in the 19th century. It was the age of science and engineering. The discovery of one law of physics or chemistry after the other gave ride to the development of unprecedented inventions. Scientific research became the battlefield of the western world. The developed nations competed for leadership in all areas of science. Huge investments built up research facilities in universities and other academic institutions. By the end of the century, science had become a byword for progress.

7It was not easy for the human sciences tocompete. History and geography found a new raison d’être in underpinning the ideological foundations of the modern nation states, to justify old or new borders and to assign to new nations a glorious past. For language studies, though, things looked different. To the extent that the national language became part of the academic syllabus in the 19th century, the focus was on literature and on the history of language. It was, like the arts, a primarily ornamental enterprise. To a large extent, the compilation of dictionaries and grammar books dealing with the contemporary language took place outside of the academic environment, organised by more or less prestigious publishers. The development of a standardised literacy did not require academic research, it therefore rapidly became a blend of commercial and administrational interests. Lexicographers and those producing classroom textbooks for teaching the national language in primary and secondary schools were seen as craftsmen, as the poor cousins of real academics. Indeed many of these authors were, or had been, teachers, and their status was hardly reflected in the immense impact they had on the normalisation and unification of the national language. They designed on the basis of existing anarchic dialects a language standard based on rules and orderliness. They streamlined the grammar, exorcised existing variants and introduced new norms. Their grammars still define for us today what a (complete) sentence is and what not. While it was the written language they standardised, they insisted that spoken language should be governed by the same rules. “Speak in complete sentences!” they exhorted pupils, and this maxim still underlies many contemporary linguistic theories, for that of Noam Chomsky. Thus the engineering of the languages of the modern overwhelmingly literate nation states took place outside of the universities. There, the philologists were seen as unreliable academics who largely resisted the nudge towards patriotic competition and preferred to co-operate across borders. Academic philology came to be seen as the unattractive, tedious toil of scrutinising mostly very dull texts, at best what is called an Orchideenfach in German: a discipline like an orchid: rare, exotic and hardly relevant.

2. Ferdinand de Saussure and the foundation of linguistics as a science

8This the was situation when Ferdinand de Saussure entered the stage. He had studied philology at German universities and was well aware of its shortcomings, the most serious of them that it defined itself as part of the human sciences. Like his contemporary Sigmund Freud in Vienna, who insisted that psychoanalysis was taking psychology into the hard sciences, like Emile Durkheim in Bordeaux and later in Paris who became a founding father of modern sociology, he reinvented language studies as modern linguistics, as a science on par with physics and chemistry. Emile Durkheim is credited with turning sociology into a science, not a natural science, but in every aspect as fact-based and systematic as any of them. As an heir to Auguste Comte, he saw society as more than an assemblage of individuals, namely as a system governed to a considerable extent by general laws that could be discovered by an unbiased, scientific observation of social facts. In Durkheim’s view, society was more than the assemblage of individual minds. People were bound, in their symbolic behaviour, by the constraints that society put on them. In order to describe society as an object of science, one had to shed all philosophical speculation about society, speculations such as proffered by Jeremy Bentham and other utilitarians, but even more so by Thomas Hobbes. For Durkheim, society was a mechanism, a clockwork that would move on governed by its own laws and not by the whims of the individual cogwheels who made it up.

9Saussure never quotes Durkheim directly, but Durkheim’s idea of society as an autonomous entity permeates the Cours de linguistique générale. For Saussure, language, la langue, is a system of communication; and therefore he views it as a social institution, a social phenomenon. As he explains in Chapter III, § 1, the linguistic faculty is not, as claimed by Paul Broca, something that we find in some area of the brain, but it is a “more general faculty governing signs” that exists “over and above the functioning of various organs.” The symbolic behaviour of humans has to be accounted for not in terms of mental or neural facts. Meaning does not occur in an individual mind in isolation; it presupposes the other, it presupposes society. The language system may be situated in people’s brains, but as society overrides individuality, one’s linguistic faculty is no more than an implant of the linguistic faculty of a society: “It is a fund accumulated by the members of the through the practice of speech, a grammatical system existing potentially in every brain, or more exactly in the brains of a group of individuals; for the language is never complete in any single individual, but exists perfectly only in the collectivity.” (Chapter III, §2) As Roy Harris, his English translator and one of his most trenchant critics, sees it, “[t]he Saussurean programme for linguistics was an ingenious and face-saving compromise. It turned on the identification of an object of study in the individual and a social institution in the community at one and the same time.” (Roy Harris (1987): The Language Machine. Ithaca: Cornell University Press, 44) For Saussure, for Durkheim, there was no compromise: for them, members of a society were never entirely free; they were bound by its institutions. In order to communicate, they had to employ language as a social system.

10For me, the problem with Saussure is his division of langage into langue and parole. He defines langue as the system, the apparatus, the mechanism, the clockwork that has to be studied by abstracting from what was actually said. While it is, more or less perfectly, internalised in the minds of the members of a language community, it is part of a society which controls what speakers can say and what they cannot say. The key elements of this language system are signs, realised as words, which are, for their users, “concrete units” (148), in spite of questions concerning their delimitation and the relationship between lemmas and word-forms. As signs, they have two aspects: signifier (form) and signified (content). The value, or meaning, of a particular sign is, according to Saussure, determined by its relationship to all the other signs, rather than by its reference to some discourse-external reality. This definition is, as I see it, a consequence of modelling langue in a strictly synchronic space, thus excluding all prior negotiations between the members of the language community that served to fix this value. It also suggests that lexical items have a fixed value of their own, in isolation, without being embedded in a context. In Saussure’s view, words in the context have no impact on its meaning. More or less fixed phrases or idioms have no place in his system. The phenomenon of collocation, this key conception of corpus linguistics, is not discussed. The meaning of a sentence is thus systematically fixed and not in need of a interpretation; it is the combination of its syntactic structure and the values of the words of which it consists; it can be calculated in the same way as for instance physicists can calculate motion. A calculation is something scientific, unlike an interpretation. An interpretation is a contingent act with an unpredictable outcome; it is the standard fare of the human sciences. Their remit is to make sense.

11La langue has us calculate the meaning of a sentence by using a scientific methodology and gives us a verifiable equation. It is something a well-trained computer can do. La parole has us interpret the same sentence in the context of a text and of an unfathomable discourse. Computers cannot do it. The langue perspective gives us a predictable formula, the parole perspective a possible paraphrase. La langue dissociates a sentence from the reality of the discourse, of that what has actually been said. But only the discourse is real. The sentence in isolation is an hypothetical, artificial entity. Abstracted from an empirical basis, it does not mean anything. Even if we take the formula as equivalent to the mental representation of the sentence, it remains meaningless, unless it can be linked to the symbolic interaction between the members of society.

12Of course it is possible to construct a model of what we take to be this language system. But to know if this model is plausible, if it helps to make sense of what has been said, it would have to be tested against the discourse, real language data that constitute la parole. The attempt to establish linguistics as a science on par with the natural sciences, within the paradigm that Durkheim had devised for his sociology, had, in the end, to be self-defeating, as the exclusion of real language data meant that there were no facts that could be scientifically observed.

3. From structuralism to cognitivsm

13Yet there can be little doubt that Saussure set the tone for 20th century linguistics. His approach gave rise to what was subsequently called structuralism, a method that allows us to sort the elements we deal with into pigeonholes, by defining the features that distinguish them from all other elements. Louis Hjelmslev and Roman Jakobson later developed into a concrete methodology what Saussure himself had only sketched. This meant that they abandoned to some extent his strong division and took steps to reintegrate la parole into la langue. It resulted in a concept of language that can be studied under three interrelated perspectives: system, norm and usage, and found its ultimate formulation in Eugenio Coseriu’s writings which also reintroduced the diachronic perspective that Saussure had excluded for the study of la langue: Sistema, norma y habla (Montevideo 1952) and Sincronía, diacronía e historia. El problema del cambio lingüístico (Montevideo 1959).  

14It would be interesting to explore the question to which extent the European structuralists (and for that matter, the American structuralists Leonard Bloomfield and Zellig Harris, as well) considered the language system they were describing as the ‘real’ linguistic faculty itself or as a model of this faculty, a useful construct that ideally would be functionally equivalent with the ‘real’ linguistic faculty. Saussure never made it clear whether he saw his observations as leading towards an accurate descriptions of la langue, or just towards a model that would be functionally more or less equivalent to this linguistic faculty. To me it is obvious that subsequent structuralists such as Jakobson, or even more so Algirdas Greimas and Bernard Pottier, aimed at developing a theory that would show how language works, but that they did not see their theory as a discovery of something that is in any sense ontologically real.

15Noam Chomsky, however, saw himself very much in the role of a discoverer of a ‘real’ language faculty, something he frequently also called a language organ. For him, this faculty is part of our genetic endowment, is indeed a human (and only human) phenotype. He situates it in the mind, not in the brain. As part of our genetically determined equipment, it must be universal, the same for every human being. It is therefore not a social phenomenon, not part of the contingent makeup of a culture or society. Regardless of which language people actually speak, they all share the same language faculty that regulates what can be said and what not. French or Chinese or English or Tzeltal are nothing but surface variations of the same underlying mechanism, generating (at least in Chomsky’s early grammar version) the same deep structures. The language faculty is a module of the mind. We must therefore assume that for him the mind is more than a convenient theoretical construct, it is something that really exists, in the same way as there are software programs in a computer. This is what puts Chomskyan linguistics into the same paradigm as (all varieties) of cognitive linguistics. They all take the existence of a mind for granted, and they all use the computer as a model for the mind. Indeed the arrival of the computer, the fascination with a mechanism that could ideally lay claim to thinking (we only have to think of the Turing machine) marks also the begin of all cognitive sciences. While few today explicitly refer to the computational theory of mind, it keeps lingering in the background of mentalism and cognitivism of all sorts. Google still lists 2500 hits for “computational theory of mind”.

16Cognitive linguistics, which now makes up the mainstream in theoretical linguistics the world over, is based on the hypothesis of an internal universal language of thought that we all use in our mental representations of content and which we, for the purpose of communication, have to encode (as speakers) or decode (as hearers) into a natural language. This is a view of meaning quite different from the philologists’ approach, but it bears a certain resemblance to Saussure’s views. Words are seen as fixed units, signs, that can be studied in isolation from each other, as elements of the language system in which a concrete form corresponds to a concrete value, a meaning or a concept, regardless of the fact that words are what also constitutes the discourse. While for Saussure the value of a word is determined by its relationship to all other words, for Chomskyans and cognitive linguists, natural language words correspond to mental concepts, concepts that we are perhaps born with or that are composed of semantic primes (reminding us of Greimas’ semes) which are part of everyone’s genetic endowment. Mentalists and cognitivists postulate a mechanism in our minds that translates natural language words automatically into mental concepts and vice versa, without us normally even being aware of them. According to these ideas, we all share a language of thought of which we are not conscious, a mechanism turning a thought expressed in this mentalese into a natural language sentence, and the natural language sentences we are exposed to into mentalese. There they are processed by a mechanism working like a computer, without us being aware of these processes. Like a computer program, it comprises ‘instructions’ or ‘commands.’ There are ‘laws’ or ‘rules’ determining the way the mind operates. We ourselves have no say in these operations. Chomskyan linguistics or cognitive linguistics is a science like physics or chemistry; it is not a Durkheimian ‘social science’ on par with the natural sciences. It does not deal with symbolic content as a contingent social artefact, as Saussure had envisaged langage) For mentalists and cognitivists, the laws of the universal language faculty and the (basic) mental concepts representing content are as real and permanent and inalterable and universal as the laws of physics.

4. Whatever happened to parole-linguistics?

17There is, however, another side to language, as Saussure knew well. There is parole, the is language as that what has been said and written; there is what I call the discourse. This was what the philologists had been dealing with. For them, the question was what an expression in a given context, in a given discourse, means. They did not believe that it is possible to formulate the rules determining the meaning of a word in such a way that it is possible to calculate exactly what this word means in a given context. For them, philology was at the core of the human sciences. It was not about discovering laws of natural systems but about interpreting content. In the view of Wilhelm Dilthey who formulated this programme roughly a century ago (contemporaneous with Saussure), interpretation (he calls it Verstehen) has to be seen as a dialogue between the interpreter and the text or text segment. (Wilhelm Dilthey (1981): Der Aufbau der geschictlichen Welt in den Gesiteswissenschaften. Frankfurt a. M.: Suhrkamp) Yet at the same time and not mentioned by Dilthey, interpretation also involves a dialogue with one’s interpretive community, as Stanley Fish describes it in his famous essay Interpreting Variorum (first published in 1967, reprinted 1980 in Is There a Text in This Class (Cambridge, Mass.: Harvard University Press, 167ff.)

18The meaning of a given text (segment), is to a large extent open. It is not the result of applying a set of defined procedures. For us, the only way to approach the meaning of a text (segment) is to interpret it. Our interpretation has to be based on discourse evidence. But how we assemble this evidence and what conclusions we draw from it will always be contingent and not predetermined. As we do not normally interpret texts for ourselves, we produce our readings generally as members of an interpretive community which has its specific conventions, and our goal is to see our own contributions accepted by other members of this community. Interpretation is, as Hans-Georg Gadamer never became tired of repeating, an art, and not a science. There is no methodology in the sense that if we follow all set of explicit instructions we will come up with the ‘true,’ the one and only interpretation. This is the gist of his magnum opus with its ironic title Wahrheit und Methode (2nd edition 1965. Tübingen: J.C.B. Mohr). Not many linguists in the 20th century focused their work on such a conceptualisation of meaning. Hermeneutics was seen by most of them as the antithesis to a ‘real’ science of meaning, good enough for old-fashioned continental philosophers, notably, besides Gadamer, his teacher Martin Heidegger, and then Jürgen Habermas and Paul Ricoeur. But hermeneutics was not alone in searching for the meaning of a text (segment) not in people’s heads but in the discourse. That meaning is something under constant, never-ending negotiation by the members of a discourse community; that there is no meaning without a dialogue, is also the lesson to be learned from Mihail Bakhtin (or his double V. N. Voloshinov). In the west his ideas came to prominence in the break-up of the traditional structuralism, which was the talk of the day in France in the sixties and the seventies. We can observe it, for instance, in the work of Michel Foucault (his discursive formations, in particular), in the conceptualisation of intertextuality by Julia Kristeva and in the Roland Barthes’ contextualisations of lexical expressions. That all the signifiants we ever encounter in the discourse never refer to anything by other, previous signifiants (and not to some discourse-external ‘mental concept’ or some formal expression or to an uninterpreted reality in the world ‘out there’ is the legacy of Jacques Derrida. But this discourse, Saussure’s parole or Chomsky’s performance could not attract scientifically-minded mainstream linguists. We must not wonder why Saussure never came around to writing his promised parole-linguistics.

5. The advent of corpus linguistics

19It was not before the sixties of last century that corpus linguistics entered the stage. The corpus is a principled collection of real language data, of texts (or parts of texts) sampling and thus representing a given discourse. From the beginning, corpus linguistics, as conceived by John Sinclair, was strictly empirical. It is quite surprising to see how the formative principles of this new paradigm have been present from the very beginning, as any close reading of the first major work in corpus linguistics, English Collocation Studies, shows. This book was written in the mid-sixties but published only 2005 (Ramesh Krishnamurthy (ed.): English Collocation Studiesby John McH. Sinclair, Susan Jones and Robert Daley. London: Continuum). Lexical items, we learn there, acquire a meaning not in isolation but only in their contexts. What they mean depends largely on their collocates, the relevant words around them, grammatical or lexical. The constitution of meaning is the result of collocation, of linking together words in a given context, repeating what has been linked together time and again in the discourse, and thus in the corpus. Corpus linguistics was originally the response to the need of teaching English as a foreign language. Traditional dictionaries, with their focus on the single word in isolation, could not tell its users how to use a word. Corpus linguistics was  an approach that could remedy this deficiency. It was not conceived not a reaction to the self-defeating endeavours of Chomskyan and cognitive linguistics to describe the way in which the individual mind processes language. It was firmly situated in what is called in the British context applied linguistics, with its focus on language teaching and dictionary making. Corpus linguistics did not see itself as an alternative or competitor to paradigms claiming to discover, or at least to model, the reality of a language-specific or a universal language faculty.

20Initially, corpus linguistics was not concerned with the interpretation of unique occurrences. It did not see itself in the tradition of hermeneutics. Corpus linguistics was developed as a methodology to find out things about a language in general, in order to help with the tasks of applied linguistics. The statistical tools that were put to use were supposed to find out what was common to all the occurrences of a lexical item. What were the other relevant words in all the contexts combined of these citations, extracted from the corpus and presented as concordance lines? Sophisticated statistical software could discover which other words were co-occurring with the lexical item in question (the node) in a significant way, significance implying they would occur more (or less) frequently that would be expected from a random distribution. Those words were the collocates of the node word, and node and collocates together would form a collocation. Corpora were compiled with the aim to describe this kind of lexical patterning. The goal was to formulate generalisations, something that could or should be taught or entered into a dictionary. The focus on what was common to all occurrences of a lexical item (or to a subset of them) or to all occurrences of a larger unit, a more or less fixed phrase. This goal required a synchronic perspective, just as such an (anti-philologist) perspective had been the basis for Saussure’s langue.

21I purposefully use here the term ‘lexical item’ instead of ‘word’ . A lexical item is a unit of meaning; it is, from a certain perspective that needs to be defined, monosemous. Lexical items often consist of two or more words, sometimes even, as in the case of certain idioms, of a full sentence. A lexical item is more than a just a random combination of words. The meaning of civilised world, as we encounter it in the media today, cannot be reduced to the meanings of civilised and world. As a lexical item, this phrase serves to distinguish ‘us’ from ‘them’, and it assigns to ‘us’ a higher status than that accorded to ‘them’. Take this example that I found on the Internet: ‘The Iranian regime is an affront to the entire civilized world.’ The meaning of this lexical item has little to do with being civilised or displaying civilised behaviour. The speaker knows that Iran is not less civilised than the America, and that the Iranian government is not less civilised than the American government, if we take as the meaning of civilised ‘polite and good-mannered.’ The phrase civilised world is a unit of meaning in its own right. As we find it f in our western media, it almost always refers to nothing but western countries, definitely the United States, often also to th United Kingdom, and sometimes also to their western allies. Some time ago, I have analysed 450 post September 11 articles containing this phrase in the London Times, and this list of collocates can be read as a digest of its current meaning, the essence of what has been and is being said about it:

The civilised world is peaceful, legitimate, dignified and solemn, has fundamental responsibilities, is full of innocent people asking for freedom, democracy, reason, and tolerance.

Its (sworn) enemies are: terrorism, (irrational) fanaticism, (global) terror, terrorists, fanatics, suicide murderers, aggressive dictators, Muslim militants, the Muslim world, religious extremists, religious hatreds, bigotry, this cancer.

The civilised world is under (ruthless) attack, assault, terrorist attacks, but when stirred to anger and to action it will wage a (titanic) struggle for freedom and security, it will bring its enemies to account, punish them and hound these people down.

22None of the larger monolingual English dictionaries lists civilised world as an item. Traditional linguists have, for a long time, overlooked the fact that our utterances are much less the result of an infinite linguistic creativity, they are endless reiterations and variations of formulaic ready-mades. Yet we are not tuned to recognise even frequently recurrent phrases as potential units of meaning unless they belong to the limited set of idioms that we recognise as our linguistic heritage. Only a statistical analysis of a large amount of real language data can reveal significant co-occurrences of single words, occurrences that signify to us that here we have not stumbled across a random combination of them but a unit of meaning so far not listed in any dictionary. Only corpus analysis makes us aware of such units of meaning. That meaning only emerges when single words co-occur with other words and are embedded in a wider context, inside a text embedded in a discourse, is something we have realised only with the advent of corpora and adequate software.

6. What is new about corpus linguistics?

23For a long time it was not seen worth a discussion whether corpus linguistics, as part of the applied linguistics spectrum, is on par with the hard sciences. It was developed to serve a purpose, to provide better descriptions of (variants of) natural languages. Corpus linguistics thus is strong on methods; and not too much has been said about its relationship to other ways of doing language studies. What distinguishes it from the other mainstream linguistic theories is its insistence on a bottom-up approach. Generativists, for instance, presuppose structures such as noun phrases and verb phrases which they see as constituents sentences, and structuralists have conceived semantic features called semes which they see as the building blocks of sememes, representations of word meanings. These are top-down categories, designed to describe the evidence before this evidence has ever been scrutinised. Corpus linguistics takes no descriptive categories for granted, neither words nor parts of speech nor the constituents of phrase structure grammar nor semantic features. It starts by investigating the evidence. What do we find as linguistic phenomena in a corpus, what elements have a tendency to co-occur, which patternings can be observed. Only then corpus linguists ask how can the data be categorised. Computational tools are the realisation of an impartial methodology. The computer does the counting, the measuring, the comparing and the sorting. The computer programs used to carry out these tasks are open to scrutiny and have to be approved by the peer community. The results they produce are reproducible; whenever the programs are running over the same corpus, identical results will be produced. This makes the methodology of corpus linguistics analogous to that of the natural sciences such as physics or chemistry. Indeed, the methodology of corpus linguistics can lay more claim on scientificity than Chomskyan or cognitive linguistics. Corpus linguistics does not deal with models of putative but inaccessible systems, it deals with real language data, data that are there for everyone to inspect.

24It is this computational methodology that makes up the advantage corpus linguistics has over its predecessor, philology. However large their filing cabinets were, however much they took care to be comprehensive and unprejudiced in their observations, philologists could not distinguish objective findings from interpretations and they never could do justice to their whole corpus. They could only deal with what caught their eyes. Yet patternings often become only visible in a statistical approach. It took the advent of the computer to demonstrate the ubiquity of collocation.

25Recently the corpus linguist Michael Stubbs has discussed in which way this new approach deals with the notion of the language system. He rejects, though not explicitly, the Chomskyan and cognitive claim to situating it in the mind. Instead Stubbs insists that the language system cannot be separated from the discourse, from real language data. What we can call a language system is what can be generalised from people’s symbolic behaviour, from the reality of peoples verbal interaction. Corpus linguistics discovers “patterns which are otherwise invisible: what is expected, predictable, usual, normal and typical in language use.” But he concedes: “We have as yet only the outlines of a theory which can relate system and use.” Michael Stubbs (2007): ‘On Texts, Corpora, and Models of Language’, in: Michael Hoey, Michaela Mahlberg, Michael Stubbs and Wolfgang Teubert: Text, Discourse and Corpora. London: Continuum, 155).

26One could say that corpus linguistics has become a victim of its own success. It is this scientific apparel that makes this methodology attractive even to linguists working in other theoretical frameworks. It supplies the basis for any theoretical negotiation that takes place within the expert community. Today, most theoretical linguists insist that they are committed to empirical data and to corpus methodology. There is even a journal called Corpus Linguistics and Linguistic Theory, claiming implicitly even in its title that corpus linguistics is not a new way to look at language but just a new bunch of methods and computational tools to access discourse evidence. But there is a key difference between corpus linguistics and other paradigms. A cognitive linguist would have developed a particular theory that is defined in categories and the relations obtaining between them, and any hypothesis they might want to test would be expressed in terms of these categories. The corpora cognitive linguists are working with come therefore usually in annotated form, the annotations, whether tagged automatically, or (semi-)manually, indicating the pre-conceived categories in question. The results of such a computational analysis therefore cannot question the validity of the categories themselves. Basically researchers using this approach look in the corpus for examples for the structures or entities they have hypothesised. This is a use of corpus linguistics that has been termed by Elena Tognini Bonelli as ‘corpus-based’, as opposed to ‘corpus-driven’ for the bottom-up approach (Elena Tognini Bonelli (2001): Corpus Linguistics at Work. Amsterdam: Benjamins, 61). Corpus linguists insist that we have to be wary of such untested presumptions, and that we have to make sure that we do not use the computer to seemingly prove their ontological existence. The notion of ‘word’ is but a theoretical construct, it is not a unit that can be detected and counted by a computer. A machine does not detect words, it only detects chains of alphabetical characters between spaces. Whether they are words is a matter of interpretation. A tagger does not identify nouns; it assigns a tag either on the basis of a list of ‘nouns’ or on the basis of some ‘training’, reproducing what are the notions of fallible human beings. There are no nouns in the sense in which there are oranges. A noun is a man-made concept. It may make sense or not. But the same is true also for the ‘significance’ assigned by statistical programs. They (and there are many of them, delivering a gamut of amazingly different results) reflect the theoretical notions of statistical significance of those who wrote these programs. These are, however, considerations that apply not only to the methodology of corpus linguistics, but also to all the methodologies we find in the hard sciences.

7. How scientific is corpus linguistics?

27While the methodology of corpus linguistics can lay claim to scientificity, the question of whether the paradigm itself can be described as scientific is more problematic. The methods, realised by computational tools, deliver results, findings that are reproducible and unassailable. But how telling are these results? The table below shows what statistics and frequency tells us when we are looking for the twenty most significant collocates of the node word heron in a large English corpus, the Bank of English with more than 500 million words. What do we really learn about the meaning of this lexical item? Significance does not mean relevance in any sense, it is measured as the deviation from random distribution. A collocate is a word co-occurring significantly in the immediate context of the node heron, defined as five words to the left and five words to the right of the node. The table below gives the first fifty ‘words’ (items between spaces), listed according to rank, that we obtain by applying the t-score and the Mutual Information (MI) test, respectively. In the third column we find the fifty most frequent contexts words, with their raw frequency:

t-score

Mutual Information

Raw frequency

patrick

scott

gil

island

s

rc

flight

blue

and*

liz

night*

great*

ronson

artist

kingfisher

chick

was*

st

crowned

gerald

painter

taylor

james

international

by*

pippa

nolan

green

abstract

mike

fish

city

<p>*

lyne

purple

stakes

painting

white

grey

ives

001

1981

tower

2*

a

</date>

corporation

lake

baby

park

79

66

51

42

149

28

28

28

276

22

27

26

17

14

13

13

93

15

11

11

11

12

13

14

64

9

9

11

9

10

10

14

122

8

8

8

8

12

8

7

7

7

7

17

218

9

7

7

8

9

8.851986

8.064827

7.136570

6.358066

5.778779

5.283019

5.193544

5.102982

4.984057

4.661392

4.482823

4.333431

4.120879

3.648895

3.597146

3.595990

3.512002

3.463732

3.305039

3.285298

3.284370

3.282829

3.278711

3.065846

3.027831

2.997651

2.982007

2.973864

2.968457

2.964786

2.913201

2.884250

2.847704

2.825549

2.785454

2.751572

2.728219

2.717118

2.710228

2.638379

2.625273

2.556060

2.552501

2.547713

2.522361

2.517748

2.506733

2.504475

2.498620

2.475114

cyclerama

aorg

triandis

6730

allyn

merganser

0749

lanyon

grebe

ronson

bishopsgate

metronome

gil

pippa

19990322

cormorant

lyne

herons

rc

dipper

snipe

sisterhood

729

crested

osprey

kingfisher

chick

ives

paragon

crowned

burghley

kempton

patrick

denton

heron

nocturnal

beak

turtles

mystique

breasted

nolan

albans

liz

scott

001

002

stevie

suzuki

headingley

cooks

6

3

4

4

5

3

4

3

5

17

4

5

51

9

4

4

8

3

28

3

3

3

3

4

3

13

13

7

3

11

3

5

79

3

4

3

3

4

3

3

9

3

22

66

7

3

4

3

3

4

15.523910

15.331246

13.576183

13.102205

12.395493

11.938590

11.679830

11.658454

11.497974

10.856020

10.690538

10.567891

10.522654

10.319519

10.094710

9.990863

9.941615

9.302739

9.285664

9.145754

9.026837

8.946905

8.942592

8.808995

8.753161

8.745627

8.559589

8.488213

8.190983

8.161973

7.997359

7.949833

7.940231

7.906339

7.749760

7.692048

7.542567

7.511884

7.497575

7.438675

7.382086

7.356036

7.337060

7.100897

7.014135

6.977266

6.951845

6.927396

6.803897

6.751075

the

and*

a

of

s

to

in

<p>*

was*

is

patrick

for

on

scott

by*

at

with

as

gil

that

his

island

from

it

but

has

one

this

an

blue

flight

rc

he

night*

or

had

all

great*

who

who

said

liz

be

are

our

which

up

when

will

2*

373  

276  

218  

171  

149  

126  

125  

122  

93  

81  

79  

77  

76  

66  

64  

60  

55  

53  

51  

50  

43  

42  

36  

32  

32  

29  

29  

28  

28  

28  

28  

28  

27  

27  

26  

26  

26  

26  

24  

24  

23  

22  

21  

21  

19  

19  

18  

17  

17  

17

28None of these lists tells us much. It is not surprising that most of top words in the raw frequency list are grammatical words. But the t-score list and the MI list are hardly more informative. Obvious t-score prioritises words with a higher frequency, while MI disregards frequency altogether. In both lists we find many ‘collocates’ that seem to make no sense at all. Some of them have to do with the coding conventions of the corpus. Others appear to be totally random.

29Collocates occurring in all three lists are underlined. Words only occurring in the first and second list are given in italics. Collocates  occurring in the first and the third list have an asterisk. Once we deselect proper names, such as Patrick, Ronson, Liz, Scott etc., there are few collocates which make it on all three lists. On the whole, the t-score list has the highest number of relevant collocates, such as night, flight, kingfisher, blue, crowned, fish, grey, lake. To recognise them as relevant we have to have some previous knowledge about herons, for instance that they can fly, that there are different varieties (crowned, blue, grey herons), that they like fish, are found near lakes and so on. The mutual information list has not much to offer. It adds cormorant to the kingfisher of the first list, both being comparable birds.

30This is how heron is described in the Cobuild English Dictionary (Cobuild English Dictionary for Advanced Learners, 3rd edition, 2001, Glasgow: Harper Collins Publishers), the first dictionary strictly based on corpus evidence and edited by John Sinclair):

A heron is a large bird which has long legs and a long beak, and which eats fish.

31It takes a lot of interpretation to jump from the results provided by the computer to this sentence. Of course, there is more to do for us corpus linguists than just search for collocates. We can also look for paraphrases, definitions, or explanations of heron, for instance for all citations of “a heron is”. These are some more relevant findings (out of 73) provided by Google:

A heron is a large wading bird in the family Ardeidae

A Heron is a bird that hangs out around the water; by the beach.

A heron is a beautiful creature.

A heron is a huge target. Hard to miss.

An image is no more a representation of an existing reality than a heron is a representation of the landscape through which it is flying.

A heron is a lonely and solitary creature often found sobbing in ladies washrooms lamenting its gothic heritage

To some, the appearance of a heron is a good omen, but to others, it warns of fire and drought.

A Heron is a bad omen, heralding parsimony and evil.

32Many other citations, contextualised occurrences of the word heron. also contain paraphrastic content of this kind. In the British National Corpus, a 100 million public domain corpus representing British English of the early 1990s, we find, among the 289 occurrences of heron, for instance these citations explicitating what can be said about this bird (highlighted in italics):

a tall stooped grey figure, like a heron

statuesque heron stared balefully into shallow pools

a heron ... would hunch on the garage roof as I left for work, and I could well imagine the thoughts of piscicide running through its reptilian brain.

This long neck and dignified strut makes the heron's posture an enduring image.

A heron scraiked below the steep slabs of stone

Forget shrill heron's stark cry

watching a heron gliding around like a poker with a crick in its neck

a heron rose on gaunt wings, its legs dangling

The old heron ... flaps his melancholy wings.

That most attractive bird, the heron can be observed more easily

33Sampling the paraphrastic content of Google, the Bank of English and the British National Corpus, we might come up with this interpretation of the meaning (copied [i.e. not inferred] paraphrastic content in italics):

A heron can be found close to the water, by the beach. It displays a tall stooped, statuesque, posture, featuring a long neck and a dignified strut. Its wings are gaunt and perhaps even melancholy. When it flies, it may look like a poker with a crick in its neck. The heron's cry can be called shrill and stark, a kind on skraiking. It lives of piscicide, and humans may disagree with its baleful intentions ascribed to its mind. Because it presents a huge target, it can easily be shot dead by those who do not like it. Others may find it a beautiful creature and a most attractive bird. Whether herons are seen as a good omen, or a bad omen, warning of fire or drought or heralding parsimony and evil, is an open question. In some cultures, a herons is ridiculed as a king among a hundred crows. More context would be needed to understand why it is said that a heron is a lonely and solitary creature often found in ladies washrooms lamenting its gothic heritage.

34My interpretation, as it stands, is fully backed up by real language data. But it is not a scientific finding. It is not the meaning of heron, not even within the discourses for which these three corpora stand. It is not the direct result of applying a scientific methodology. Other people working with the same corpora would have different ideas which of the collocates are relevant (as opposed to significant), they would choose other citations, they would also have different ideas of what paraphrastic content is. The meaning of heron within a discourse, or within a corpus, is all what is being said about heron. But what is all? Meaning, as I see it, does not distinguish between word knowledge and world knowledge, between lexical and encyclopaedic knowledge. The computer cannot ‘decide’ for us what counts as such knowledge; such decision-making remains a conscious act, and as such, it is contingent, subjective and never final. I can offer my interpretation to the interpretive community of which I am a member. There it will be negotiated, in yet more interpretive acts. Even if we should all, in the end, agree to one interpretation, this interpretation could not be called scientific or reproducible. The same interpretive community, sighting the same evidence tomorrow, might well arrive at a different outcome.  

35Corpus linguistics, as I see it, can only partly be seen as hard science. Its methodology and its computational tools may well deliver dependable and reproducible results. But these results do not tell us much. They need to be interpreted. Yet this step takes corpus linguistics into the human, the interpretive sciences.

8. How scientific is cognitive linguistics?

36The task of the cognitive linguist, as seen by Ronald Langacker, is to provide structural analyses and explicit descriptions of thoughts and concepts. Langacker would not be a cognitive linguist if he did not place meaning quite firmly into the realm of the mind. For him, “[m]eaning is equated with conceptualisation. Linguistic semantics must therefore attempt the structural analysis and explicit description of abstract entities like thoughts and concepts. ... Because conceptualisation resides in cognitive processing, our ultimate objective must be to characterize the types of cognitive events whose occurrence constitutes a given mental experience.” (Ronald W. Langacker (2002): The Cognitive Basis of Grammar. Berlin: Mouton de Gruyter, 2) But he keeps an open mind: “Could it be, for example, that language resides entirely in individual minds? ... At the other extreme, it might be maintained – perhaps from a strictly Platonist or interactionist perspective – that language has no cognitive representation whatsoever.” (Ronald W. Langacker (1997): ‘The contextual basis of cognitive semantics’. In: Jan Nuyts and Eric Pedersen (eds.): Language and Conceptualization. Cambridge: Cambridge University Press, 229-252, 229) It is through:

interaction and mutual accommodation that the conventional patterns of activity we call 'language' arise, sustain themselves, and undergo continual adjustment. The study of its interactive basis is therefore essential to a full account of the acquisition, structure, and use of language. I insist, however, that it is not equivalent to such an account, which must also encompass the specific mental adjustments that enable speakers to communicate in a conventionally sanctioned manner. To focus exclusively on the interactions per se would be as pointless as ignoring them altogether – either option omits an essential component of language, neither of which can be well understood without the other. (233/4)

37At the end of the day, however, it is meaning we are concerned with. Once conventions, mutually agreed rules, are obeyed, we must not shy away from looking into people's heads: “The meanings of linguistic expressions are conceptualizations, and the relevant conceptualizers are the speaker and the addressee.” (242)

38Meaning, then, is what we have to look out for in the depth of the mind, the way an “expression is actually comprehended – its full, richly detailed contextual understanding.” (235) “The immediate goal of semantics, for example, is to identify conceptual components and elucidate the structure of the complex conceptualizations evoked by linguistic expressions.” (240) . This is about as far as Langacker takes us. What we are not given is an illustration of such conceptualisations and the conceptual components they consist of. What is a “full, richly detailed, contextual understanding” of an utterance? This is what we are not told. No example of a “cognitive representation” is given. Instead, we find the escape route habitually taken by cognitive scientists when the readers expect a glimpse of mental activity: the sudden jump from the mind to the brain:

39Now, I do talk about mental representations, but the term does not refer to complex depictive entities stored as such. It merely indicates the occurrence of neurological adjustments, notably changes in the strength of synaptic connections that influence subsequent processing and facilitate the emergence of patterns of activity constitutive of particular kinds of mental experience. A speaker obviously has no introspective access to these adjustments, which do not in any case resemble the emergent experience they engender. In fact, even that experience lies beyond our reach, in the sense that we have no conscious access to the character of the cognitive processing that constitutes it. Although to some extent we monitor our experience, and may even attempt to analyse it, for the most part we merely have it. (239)

40Thus, I fear, the “full, richly detailed, contextual understanding” of an utterance is an experience that we have but cannot really be aware of. No introspection allows us to observe our mental states, our own intentionality. There is only one escape alley: Mental representations can best be described as neurological adjustments. The experience itself remains ineffable. As their last resort, cognitive linguists of all feathers seem to agree that it is only the neural sciences that can enlighten us about the meaning of words, phrases and utterances. Semantics has to “elucidate the structure of the complex conceptualizations evoked by linguistic expressions.” But in the end, these representations do not have to be visited in the mind but have to be understood as “neurological adjustments” of which the language user is unconscious. Like other cognitive linguists and philosophers of mind, Langacker, too, fails to come up with a single example of a mental concept.

41Does neurolinguistics, becoming more and more popular these days, offer the answers that cognitive linguistics cannot provide? Brain studies, as part of biology, certainly belong to the hard sciences. Can they enlighten us about the ‘real’ meaning of the word heron? Have we found it once we know which neurons are firing and which synaptic connections are energised whenever someone utters or hears heron? What is the relationship between a mental concept and neurons? We are still left with the key question that has been troubling western philosophy ever since Descartes threw it up: the relationship between body and mind.

42Language is symbolic. It consists of lexical items that have a meaning. Symbols are forms standing for content. But what is this content? All signs presuppose people and presuppose a dialogue. I can only know that a lexical item means something to someone (including myself) if I am told that it is a sign and which meaning it has. It is the discursive negotiation of their content that turns something, phonetic sequences for example, into signs. In a world without symbolic interaction there are no signs. Without people who know and can argue what signs stand for there are no signs. If mental concepts neural phenomena are entities of which we are not conscious, then they cannot be signs. They do not signify anything to anyone, not even to neurons. Neurons cannot argue if they should fire or not if there is a phonetic sequence perceived by the brain. Neurons undoubtedly are objects of scientific enquiry. But there is no bridge whatsoever that would allow us to connect firing neurons to meaning. Meaning cannot be but social. Meaning is neither in the mind nor in the brain.

9. Intentionality and the discourse as the collective mind

43Cognitive linguistics looks at language from a mental, a psychological perspective. But language presupposes society. Cut off from communication, children do not acquire language. However we, as the ‘symbolic species’, to quote the title of Terrence W. Deacon’s 1997 book, are genetically primed to communicate by symbolic interaction, to realise that there are objects standing for arbitrary content is something that has to be learned through interaction. Signs, words for instance, signify something to us. We are aware of their meaning. We know what the graphemic sequence heron is about. We can tell people what it means, we can remember it and forget it. Human beings are intentional. They do not only behave, they can make themselves aware of what they are doing, what they have been listening to and what they have just said. Ants, cats, dolphins, primates are also said to communicate. But they are not aware of the fact that they are communicating. They cannot discuss whether what they did was communicating or not. They do not possess intentionality. Neither do mental or neural processes of which one is unaware possess intentionality.

44Ants respond to sense data, but they do not interpret them. It is the self-referentiality of the discourse, the possibility to consider and to interpret what has been said, that is responsible for the emergence of intentionality, of consciousness. Our dexterity to discuss what we are told, to deliberate, to ponder, to know what we are thinking, to reflect about content individually and collectively is not something that we are born with; it is a capacity we develop only by engaging in a dialogue with other members of our discourse community. Our internal dialogue is but a dull mirror of the dialogue that goes on in the discourse. Without verbal input from the outside we may still operate on some level, but we do not contemplate over content. The monadic individual mind is mute. As Mihail Bakhtin said a long time ago, it is the heteroglossia of the dialogue that turns us from automatons into creative actors.

45Meaning exists only in the discourse, in la parole. Saussurean, Chomskyan and cognitive linguistics had hoped to be able to deal with meaning by focusing on the language system. But for a language system that is modelled independently of the discourse, meaning is some fixed property that comes with form. The meaning of heron would not available for negotiation; we, the speakers, would have no say in what it means. It would be an invariable formula, like the ardea in the taxonomy of Linnaeus and nothing else. That some people find the bird attractive and that anglers may see it as their enemy would not be part of its meaning. It is a kind of language description as fascinating as logarithmic tables or as aeroplane maintenance manuals. It is a language of whose meaning we would not need to be aware, for we would not have a say in it; we could not change it. A mechanism, like a computer, defined as a set of mental or neural processes, could deal with it. The concept of meaning we find in 20th century mainstream linguistic theories does not have to embrace intentionality. Their language systems would work quite as well or even better if they were no speakers at all.

46Corpus linguistics looks at language from a social perspective. It is the linguistics not of the language system but of la parole, of the discourse. Yet the more we accept this view, the less we can see corpus linguistics as a hard science. Ant researchers can study the communication system of their ants from the outside. They occupy an Archimedean point. They can assign functions to the pheromones exuded by ants without asking them. The ants have no say in the communication system that the ant researchers model. Ants have no intentionality. They do not know that they are communicating. An ant researcher does not depend on their approval. As long as the methods are sufficiently spelt out, as long as the data is made available and the experiments can be reproduced, other ant researchers will agree with the findings. For corpus linguists, the situation is quite different. They must interpret the results provided by their methodology. They must negotiate their interpretation with their fellow corpus linguists. But their interpretive community is wider. It also includes the community whose discourse they are analysing.

47Meaning is social. Therefore we must study it in the discourse, not in the minds of individuals. The aboutness of a text (segment) is something that is negotiated in dialogic interaction. Intentionality can be seen as a social, not just as a mental phenomenon; it becomes observable once we look at it as the emergent property of a discourse. How it affects an individual mind is not within the remit of linguistics. An approach to language from a social perspective will analyse a discourse as a collective mind. The meaning of a contribution to the discourse is equivalent to the traces it leaves in subsequent text, to the impact it has on the discourse.

48Corpus linguistics interested in meaning thus has to move on from the role it has played over the last forty years. As part of applied linguistics, it uses an innovative bottom-up methodology to revise the generalisations that traditional language studies, working with a canon of received language categories, have supplied. But it was geared towards the kind of linguistic knowledge that is needed for language applications such as language teaching and lexicography. Here it has brought about unparalleled innovation.. All recent dictionaries claim to be based on corpus evidence. Language teaching has been revolutionised by the use of corpora.

49But generalisation alone can never capture the phenomenon of meaning. Meaning is never finite. The discourse consists of layer after layer of interpretation. Each new negotiation of meaning, each new interpretation adds to the meaning of what has already been said. The discourse has a diachronic dimension. This is where I see the future of corpus linguistics. Corpus linguists have learned to capture what is common to all occurrences of a lexical item in a synchronic corpus. Now they must learn to find out what distinguishes a single occurrence from all the previous and all the subsequent occurrences. The must develop methods that will tell them what makes a single occurrence unique.

10. Corpus linguistics and hermeneutics

50Whatever we may say, we never start at point zero. The texts which make up a discourse are not entered into it simultaneously. Whenever a new text is being contributed, we have to understand it as a reaction to something that has been said before, and as a request to comment on it subsequently. In the text Marxism and the Philosophy of Language, which claims Valentin Voloshinov as its author but was probably conceived by Mikhail Bakhtin, we find these sentences:

[T]he monologic utterance is, after all, already an abstraction. …Any utterance – the finished, written utterance not excepted – makes response to something and is calculated to be responded to in turn. It is but a link in a continuous chain of speech performances. (V.N. Voloshinov (1989): Marxism and the Philosophy of Language. Cambridge, Mass.: Harvard University Press, 72)

51Each new text repeats to a considerable extent things that have already been said. Text segments, collocations, complex and simple lexical items which have been used previously, are being recombined, permuted and embedded in new contexts, containing some new together with old paraphrastic content. A new text repeats, accepts, comments, modifies, reinterprets or rejects what is contained in previous texts. Most of what we say is a reformulation of things sad before, but normally each utterance brings in one little innovation. It is this innovation which makes it a unique occurrence, something that in some aspects differs from all previous and subsequent occurrences.

52Therefore it makes sense to look at the discourse as a sequence of textual contributions. Earlier texts leave their traces in subsequent texts, and later texts refer to previous texts, sometimes explicitly, but more often implicitly. It is the phenomenon of intertextuality that provides the links between the texts. The word and the concept of intertextuality is relatively recent. It is attributed to Julia Kristeva, who introduces it in her exegesis of Bakhtin’s terms dialogue and ambivalence: “[A]ny text is constructed as a mosaic of quotations; any text is the absorption and transformation of another. The notion of intertextuality replaces that of intersubjectivity.” (Kristeva 1986, 39, Kristeva’s emphasis) The wider context shows that what Kristeva is describing here is not so much the discourse as a process of the individual mind which, in the perception of intertextual links, engenders a communication between an author and his or her reader. Since the introduction of the new term, it has been used in different ways. “Intertextuality, as a concept, has a (short) history of different articulations which reflect the distinct historical situations out of which it has emerged.” (Graham Allan, (2000): Intertextuality. London: Routledge, 58) As I have, as a corpus linguist, no knowledge of intertextuality as a mental phenomenon, I exclude the psychological reading of this concept. For in some sense, intertextuality is directly observable. Explicit quotations and obvious cases of plagiarism are good examples. In a wider view, intertextuality implies explicit or implicit links between texts or text segments of a discourse. By referring to what has been said before, the function of intertextuality adds a new layer of interpretation to something that has been said before. My interpretive community can now negotiate to which extent what I have said is acceptable as a new interpretation. This means that the concept of intertextuality is arbitrary. While new corpus tools will help us to discover candidates for intertextual links, the decision whether something is an intertextual link rests with us. There is no formal mechanism that tells us if to texts are intertextually linked. Similarities can after all be accidental. What is presented as intertextual links has to be endorsed by the interpretive community. Take as an example this sentence found in the San Francisco Sentinel on 02.02.2009: “Confidence in Britain’s ability to overcome the economic turmoil is sinking by the day, as evidenced by the precipitous decline of the pound, which has almost reached parity with the euro.” For weeks later, on 28.02.2009, we find in the Times this sentence: “Unlike Britain, [Portugal, Ireland Greece and Spain] have been unable to devalue their currency.” Is there a link between the “precipitous fall of the pound” and the ability “to devalue” one’s currency? Intertextuality is established not by a computer but by interpretation.

53I would like to develop corpus linguistics in such a way that it helps us to find out what a text, a text segment or a lexical item means in a specific context, at a given time. As Hans-Georg Gadamer has shown, such an endeavour can never be successful. If I hear Bach’s music today, I cannot help to hear it differently from the way his contemporaries heard it. What was revolutionary then has become, after Haydn, Mozart and Beethoven, the accepted tradition. But Bach’s music certainly has been influenced by other Baroque composers. There are intertextual links that connect his scores to those of his forerunners and his successors. The same holds for the discourse. The meaning of any lexical item in a new contribution is every previous contextualised occurrence, i.e. everything that has been said about it, up to the moment in which this contribution is entered into the discourse. More specifically, it is the entirety of the paraphrastic content that we find for this lexical item. This definition implies that it does not matter whether the members of the discourse community are aware of its meaning. Indeed, due to the limitations in accessing all the relevant data, the full meaning can never be ascertained. This is a definition of meaning that takes refuge neither to mental representations nor to the real world.

54Interpreting a text, a text segment or a lexical item is different from understanding it. In my private understanding of a poem, I appropriate it as a first-person experience. I will never be able to describe to someone else what it means for me. Only by giving verbal testimony of this experience, I can go the step from understanding towards interpreting. Understanding is something that takes place in an individual mind. Understanding does not require verbalisation. An interpretation is always expressed verbally and directed at others who have their say in it. Thus it is a collaborative act involving the other members of my interpretive community.

55Hermeneutics, the art (or craft, as some see it) of making sense of texts, is concerned with written, with recorded language. So is, unsurprisingly, linguistics, as both share the same remit. Even where speech is credited as the primordial form of language, linguists, for instance Saussure, have had to work with transcripts. Wilhelm Dilthey explicitly states that hermeneutics is about the “interpretation of written testimonies of human life.” (quoted in Mueller-Vollmer, Kurt (ed.) (1986): The Hermeneutics Reader. Oxford: Blackwell, 27) This is hardly surprising. Written texts (with few exceptions) do not provide us with the option of asking their author about their intentions. In the words of Paul Ricoeur, “writing renders the text autonomous with respect to the intention of the author… The relation between writing and reading is no longer a particular case of the relation between speaking and hearing.” (Ricoeur, Paul (1981): Hermeneutics and the Human Sciences. Edited and translated by John B. T. Thompson. Cambridge: Cambridge University Press, 139) In the 18th and 19th century, however, under the impact of invention of man as an autonomous, unique person and the ensuing psychologisation of human relations, writing was understood as a gateway to the author’s inner self. Immanuel Kant apparently was the first to demand that the goal of hermeneutic analysis should be “to understand the author better than he had understood himself.” (quoted in Fritz Hermanns, (2002): ‚Linguistische Hermeneutik’, in: Angelika Linke, Hanspeter Ortner, Paul R. Portmann-Tselikas (eds. ): Sprache und mehr: Ansichten einer Linguistik der sprachlichen Praxis, Tübingen: Niemeyer, 125-163, 139) This was taken up famously by Friedrich Schleiermacher who defined the hermeneutic task as “to understand an utterance first as well and then better than its author” (quoted in Frank, Manfred (ed.) (1977): F. D. E. Schleiermacher, Hermeneutik und Kritik. Frankfurt am Main: Suhrkamp, 94) and finally by the philologist August Boeckh: “The interpreter consequently understands the author better than the author understands himself.” (Mueller-Vollmer 1986, 138)

56Hermeneutics today, however, is less concerned with the author’s mind than with the meaning of a text, a text segment or a lexical item. This meaning has little to do with the author’s intentions. About them, we know nothing, we only have his or her words for them, which again may have their own intentions. The remit of hermeneutics is the discourse. For its interpretation, it does not matter whether the book quoted above, Marxism and the Philosophy of Language, was indeed written by Volshinov or by Bakhtin. Michel Foucault, Roland Barthes and many others have marginalised the role of the author. Likewise, Paul Ricoeur, referring to Hans-Georg Gadamer’s Truth and Method, states that writing “renders the text autonomous with respect to the intention of the author” (Ricoeur 1981, 139), and claims: “What the text signifies no longer coincides with what the author meant.” (Ricoeur 1981, 91)

57Yet how does hermeneutics work? If we want to know what any old text, say Plato’s Republic, or any text segment, any key lexical item in it, means, it is not sufficient to read it on the horizon of our contemporary conceptualisation. Rather, we have to interpret it in the context of other texts from roughly the same period, texts to which the Republic is a reaction, and texts which referred to it. But this is not enough. We also have to make ourselves aware of to what extent our reading of the text is indebted to what has taken place since. We have to realise how subsequent texts have influenced our perception. There is no straightforward way to regain the freshness of the experience that Plato’s work had in its time. The only way to approximate the original meaning is by making ourselves conscious of the traces Plato’s text had ever since. We find these traces in form of explicit or implicit quotes, comments and many fully fledged interpretations called commentaries. Yet we also have to take into account all the ideas that were invented later and form an integral part of our ways to think about society. Can we imagine a time when the concept of equality or of human rights was not yet available?

58The community interpreting Plato’s Republic thus depends on evidence, evidence taken from the discourse. Only a linguistic framework that is capable of dealing with it, with la parole, is in a position to provide such data. Corpus linguistics and hermeneutics complement each other. Corpus linguists provide the methodology that assures us that no piece of potential has been overlooked; and hermeneutics then gives us, the interpretive community, the level of reflection to make new sense of what has been already an object of innumerable interpretations.

59Let us look at a concrete example. In the very first sentence of the Republic we encounter the Greek verb προσευχομαι. It still exists in modern Greek and is usually translated into English as to pray. But what might προσευχομαι have meant in Plato’s days? What we call praying today is some special kind of a very personal, bilateral though asymmetric communication with God. Was this also true for Plato’s times? Or were prayers then the verbal equivalence to the sacrifices that people offered to their gods? Then it would rather be a unilateral piece of communication. The deity would not be expected to respond, even if the sacrifice was intended to generate goodwill on her or his part. Plato’s text in itself does not provide the answer. It takes up, without comment or paraphrase, what was then generally understood as the concept of προσευχομαι .Yet this concept was a result of negotiations among the general ancient Greek discourse community. There is an abundance of classical Greek texts dealing with it. Experts have scanned them time and again in order to come up with interpretations of the old concept. The many translations we find of this book take these interpretations into account. An analysis of a corpus comprising twelve translations of the Republic comes up with these equivalents of προσευχομαι: make my prayers, offer up my prayers, offer up prayer, say a prayer, pray, pay my devotions, pay my devoirs, worship. All of them can be read as paraphrases which try to bridge the abyss between what προσευχομαι meant then and what praying is today. These translation equivalents are interpretations relating the ancient concept to our modern understanding. Hermeneutics makes us aware that any interpretation of the ancient Greek word will have to take into account also the present-day concept of praying. This is Gadamer’s key message: “The horizon of the present cannot be formed without the past… Understanding [what I call interpreting] is always the process of fusing these supposedly different horizons.” (Hans-Georg Gadamer (1965): Wahrheit und Methode. 2nd ed. Tübingen: J.C.B. Mohr, 289, my translation)  The famous hermeneutic circle should be seen as an infinite spiral that allows us, an interpretive community of today, to approximate the meaning of something that has been said before, on the basis of evidence provided by corpus linguistics.

60The hermeneutic approach to meaning, as I have sketched it here, forsakes all hope for meaning as something stable or systematic. There is no universal, timeless, language-independent set of concepts, innate in our minds or carved into the granite of some metaphysical semantic ontology. Hermeneutics is not about la langue, it is about la parole. It does not attempt to describe the meaning of a lexical expression in all its occurrences, but to interpret a single occurrence by elucidating the intertextual links tying it to previous and subsequent discourse events. Parole-linguistics is diachronic linguistics. In this perspective, meaning is never final, it is always provisional. Each new occurrence of a lexical item in a text, each new reference to a text, adds something to its meaning. This may let look parole-linguistics less satisfactory than langue-linguistics. But if our social reality is a construct of the discourse it is only parole-linguistics that allows us to access it.

11. The linguist and the interpretive community

61Neither linguists nor lexicographers are experts for meaning. Their task is to be faithful keepers of the minutes, of what is negotiated in terms of meaning, between the members of a discourse community. They can show how lexical items are used and paraphrased. They can point to possible intertextual links based on surface similarities between texts. But we have not authorised linguists to tell us what is right and what is wrong. If people want to pay respect to the normalising intentions of the Académie française, they are welcome to do so. But no one can force them. Corpus linguistics teaches us that there is no ‘true’ or ‘correct’ interpretation of a lexical item, a text segment or a text. The meaning of a lexical item is all that has been said about it in the discourse. We can find it in the negotiations between the members of a discourse community. There may be disagreement, and there may be discord. Pluralism is the hallmark of any democratic society. Only a totalitarian regime would like to enforce a uniform language use. Discourse communities can discuss whether one should distinguish between globalisation and mondialisation, and any contributor to the discourse may offer their preferred paraphrase. If he or she cannot convince their fellow members and therefore remains alone in their interpretation, then this new paraphrase will leave no traces in subsequent texts. Therefore it is in the interest of each of us to use and paraphrase lexical items in such a way that others find it attractive to refer to these usages.

62The discourse is, in principle, a democratic system. Everybody has, in principle, the inalienable right to contribute to it. Everybody has, in principle, the right to be heard. Whether globalisation means ‘oil and war’, and mondialisation means ‘co-operation in  building a worthwhile future’ is not a matter of truth but of persuasion. What we say does not refer to a reality outside the discourse, it is a reaction to what has already been said. That we tend to think differently, that we take language to be a mirror of reality, is an almost innate conviction going back to times immemorial when language was no more than an extension of ostentation, of pointing to things. Most of what we talk about today is not something to which we can point our finger. Globalisation or mondialisation are objects solely of our discourse, not of some reality out there. We talk not about reality but about what others have said about reality.

63Corpus linguists look at the discourse as an autopoietic, self-referential system. That does not mean that it is not somehow connected to the individual intentionalities of its contributors. But these intentionalities are irrelevant. Not what we think matters, but what we do, our symbolic behaviour, our contributions to the discourse. These contributions turn the discourse into our collective mind, without ever infringing our freedom as individuals to say what we want. The task of the corpus linguist is to help us to make sense of the reality that we have constructed for ourselves within the discourse by extracting, organising and presenting the evidence. Yet they certainly have no monopoly on interpretation. We all are members, equal in status and equal in rights, of the global human interpretive community. Linguistics, corpus linguistics in particular, should be a programme of liberation, not of enforcing meaning.

12. Does corpus linguistics have a future?

64Over the summer of 2008, there was a drawn-out discussion about the nature of corpus linguistics on a weblist called the corpora-list (corpora-bounces@uob.no), with scores of contributions from all corners. You can find them if you access the corpora-list and search for bootcamp. Interestingly the majority of submissions did not come from corpus linguists working in the tradition of John Sinclair but from people working in theoretical or cognitive linguistics, in natural language processing, in machine translation and in artificial intelligence. It is not that these people have stopped modelling their language systems and have turned to the discourse. What has changed is that it has become a must to test one’s theories against the data. These contributors still see linguistics as a science, perhaps even as a science whose results can now be turned into the engineering of marketable applications, in the same way as the hard sciences have provided the basis for automobile engineering. Language engineering needs enormous amounts of corpora for testing applications. But the first results are there for us to use them. These days many of us already have been helped by machine translation tools. They may be still far from being perfect, but they are better than not understanding anything. Indeed it has been argued that texts written in a tightly controlled language full of terminology, such as aeroplane maintenance manuals, can better be translated by computers than by human translators. Language engineering is still in its infancy. They will have a growing impact on what can be said and what not. Today spelling checkers tell us if our texts conform to orthographical norms. Ten years from now, there may well be tools controlling the content of what we write. There are already increasingly sophisticated internet filters that make it impossible to access webpages with certain forms of subversive content. In this sense, the scientific methodology of corpus linguistics certainly has a bright future. Language engineering will increasingly control our discourse. Language experts will make sure of that.

65But the more our societies are subjected to the decisions of experts, the less convinced many of us become that the decisions they make are based on incontrovertible facts. Today, there is a growing number of philosophers of science, for instance Steve Fuller, Philip Kitcher, Bruno Latour, Helen Longino or Jonathan Potter, argue that all factuality is in the end based on interpretation. What the entities and categories are that define a scientific discipline, which methods are considered objective to detect, count and measure, and what the tools are that we accept to carry out these procedures, is something that has to be negotiated by the peer community. The new paradigm of social epistemology accepts that at the bottom of truth and factuality there is always interpretation: the negotiation of what has been said. Whether it is quantum mechanics or evolution or globalisation: the only reality that is accessible to us is the reality of the discourse, is what has been said about these ideas. All that we know is something that has been said in the discourse. We make sense of our environment by collaboratively constructing a reality that we accept.

66A linguistics independent of the discourse is meaningless. So far, corpus linguistics is the only approach that can lay claim to being a linguistique de la parole.

Haut de page

Bibliographie

Allan, G. (2000), Intertextuality. London: Routledge.

Coseriu, E. (1952), Sistema, norma y habla, Montevideo.

Coseriu, E. (1959), Sincronía, diacronía e historia. El problema del cambio lingüístico, Montevideo.

Deacon, T.W. (1997), The Symbolic Species: The Co-evolution of Language and the Brain. New York: Norton.

Dilthey, W. (1981), Der Aufbau der geschictlichen Welt in den Gesiteswissenschaften. Frankfurt am Main: Suhrkamp.

Fish, S. (1967), Interpreting Variorum, réimprimé in Fish, S. (1980), Is There a Text in This Class, Cambridge, Mass.: Harvard University Press.

Gadamer, H.-G. (1960), Warheit und Methode. Grundzüge einer philosophischen Hermeneutik, (2nd Edition 1965). Tübingen: J.C.B. Mohr.

Harris, R. (1983), F. de Saussure: Course in General Linguistics, London: Duckworth.

Harris, R. (1987), The Language Machine. Ithaca: Cornell University Press.

Hermanns, F. (2002), “Linguistische Hermeneutik”, in Linke, A., Ortner, H., & Portmann-Tselikas, P.R. (Eds.), Sprache und mehr: Ansichten einer Linguistik der sprachlichen Praxis, Tübingen: Niemeyer.

Herder, J.G. (1770), Abhandlung über den Ursprung der Sprache.

Langacker, R.W. (1997), “The contextual basis of cognitive semantics” In Nuyts, J. & Pedersen, E. (Eds.), Language and Conceptualization. Cambridge: Cambridge University Press.

Langacker, R.W. (2002), The Cognitive Basis of Grammar. Berlin: Mouton de Gruyter.

Manfred, F. (Ed.) (1977), F.D.E. Schleiermacher, Hermeneutik und Kritik. Frankfurt am Main: Suhrkamp.

Moi, T. (Ed.) (1986), The Kristeva Reader, New York: Columbia University Press.

Mueller-Vollmer, K. (Ed.) (1986), The Hermeneutics Reader: Texts of the German Tradition from the Enlightenment to the Present, New York: Blackwell Publishers.

Ricoeur, P. (1981), Hermeneutics and the Human Sciences. (Edité et traduit par John B. Thompson) Cambridge: Cambridge University Press.

Rousseau, J.-J. (1871), Essai sur l’origine des langues.

Said, E.W. (1981), Orientalism, London: Penguin.

Sinclair, J., Jones, S. & Daley, R. (2005), English Collocation Studies, in Krishnamurthy, R. (ed.), English Collocation Studies, London: Continuum.

Stubbs, M. (2007), “On Texts, Corpora, and Models of Language”, in Hoey, M. Mahlberg, M., Stubbs, M. & Teubert, W., Text, Discourse and Corpora. London: Continuum.

Tognini Bonelli, E. (2001), Corpus Linguistics at Work. Amsterdam: Benjamins.

Tyler, E.B. (1871), Primitive Culture. London: J. Murray

Voloshinov, V.N. (1989), Marxism and the Philosophy of Language. Cambridge, Mass.: Harvard University Press.

Haut de page

Pour citer cet article

Référence électronique

Wolfgang Teubert, « Corpus Linguistics: An Alternative »Semen [En ligne], 27 | 2009, mis en ligne le 10 décembre 2010, consulté le 03 mars 2024. URL : http://journals.openedition.org/semen/8912 ; DOI : https://doi.org/10.4000/semen.8912

Haut de page

Droits d’auteur

CC-BY-4.0

Le texte seul est utilisable sous licence CC BY 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.

Haut de page
Search OpenEdition Search

You will be redirected to OpenEdition Search