Recent milestones in the lexicon-encyclopedia debate

Recent milestones in the lexicon-encyclopedia debate

Peeters, Bert (2000). “Setting the scene. Recent milestones in the lexicon-encyclopedia debate”.

The lexicon – encyclopedia interface

(Current research in the semantics / pragmatics interface, 5; Bert Peeters, ed.). Oxford: Elsevier Science. 1-52.

Setting the scene:

Some recent milestones in the

lexicon-encyclopedia debate[1]

Bert Peeters

University of Tasmania, School of English and European Languages and

Literatures, GPO Box 252-82, Hobart TAS 7001, Australia

E-mail: Bert.Peeters@utas.edu.au

0. Introduction

Questions about the exact nature of linguistic as opposed to non-linguistic knowledge have been asked for as long as humans have studied language, be it as linguists, philosophers, psychologists, language teachers, semioticians, cognitive scientists, whatever. The distinction has been maintained and defended by some, attacked and abandoned by others. Those who have maintained and defended the distinction have drawn the line in many different ways. Very solid arguments have been advanced in both camps; in the course of a) editing this volume, and b) doing the research leading up to the present paper, I have seen so many that in the end, no longer sure of my own thoughts on the matter, I had to reluctantly decide to cancel my own contribution, or at least to hold it over until I would have had enough time for further reflexion.

I had intended to look at the sort of knowledge that is required for the exact understanding of phrases involving the verb begin immediately followed by a direct object (i.e. phrases of the type begin a book). I had intended to use French as my target language, since that is the field in which I do all my teaching and a sizeable proportion of my research. I need more time to figure out whether native speakers rely on linguistic (lexical) or non-linguistic (encyclopedic) knowledge to determine that an utterance such as Jean a commencé un nouveau livre is most likely to refer to a reading or a writing event, although various other events are by no means excluded. I need more time to figure out how they know that a certain set of interpretations (e.g. wounding, drinking) are once and for all impossible. The same questions have been asked – and answered in more ways than one – by various scholars, especially in the last ten years, even though none has framed them in terms of the lexicon-encyclopedia debate. That was going to be my contribution. It would of course have been an appropriate addition to this volume, but for now at least it has not seen the light of day.[2]

It has been said that, as a result of the spread of prototype semantics and the increased reliance on descriptive tools such as frames, scenes, and scripts, the distinction between lexical (or linguistic) and encyclopedic (or world) knowledge is now defunct.[3] A significant majority of the papers brought together here show that nothing is further removed from the truth. They also illustrate the inevitable fact, referred to above, that the distinction, when made, is not being made along the same lines by everyone with an interest in the matter. The present volume does not aim at resolving the issue: it is most unlikely that any issue in linguistics will ever be resolved to everyone’s satisfaction. The various contributors to this volume toe very different lines. Some work within frameworks (e.g. Cognitive Linguistics, Word Grammar) that reject the distinction; they either advance reasons why it should not be made or show through detailed case studies how much can be gained from not making it. Others were trained in the generative tradition (more particularly in Distributed Morphology), and demonstrate how a very theory-specific split between a list of so-called vocabulary items (the “Vocabulary”) and a list of meanings (the “Encyclopedia”) may clarify particular issues. Most are active in areas (e.g. pragmatics, language teaching, machine translation) where, for practical (but typically very different) reasons, one or the other distinction, more radical in some cases than in others, appears to be mandatory.

If so many people, all of a sudden, can come up with good reasons for a split, why is it that the view according to which the distinction is no longer made in contemporary linguistics could be upheld in the first place? My own view is that it has a lot to do with the fact that those who argue against a distinction have been increasingly vocal over the last few years, so much so that the opposite view was typically drowned in a sea of adverse opinion. The most vocal camp has been that of Cognitive Linguistics, an approach to language and linguistics looked at in section 1. Section 2 compares older writings of George Lakoff and Charles Fillmore, both currently associated with the Cognitive Linguistics movement, with more recent output. Section 3 identifies Haiman (1980) as the preferred source for the view that has been predominant in Cognitive Linguistics, and looks at responses by William Frawley and especially Anna Wierzbicka. Victor Raskin’s contribution to the debate is revisited in section 4. Section 5 shows some striking similarities between selected aspects of James Pustejovsky’s Generative Lexicon model and earlier work by scholars who took a stand on the issue of lexical vs. encyclopedic knowledge. Section 6 looks at the disappearance of the lexicon in Distributed Morphology and the subsequent inflation of the encyclopedic component. Section 7 returns to the past, paying particular attention to Katz & Fodor’s “Structure of a semantic theory” (1963) as well as to some of the waves it made among linguists and philosophers. Section 8 closes the loop and brings us back to the Cognitive Linguistics camp – or rather, to a famous precursor who does not appear to have been recognised as such, at least not among (most) linguists, namely Umberto Eco. Summaries of the papers included in the present volume follow in section 9.[4]

1. Cognitive Linguistics

I must disagree with Goddard (1998:15), who says that it is a “minority view, but an important (and perhaps ascendant) one, [that] denies the existence of any boundary between real-world and linguistic knowledge”. Goddard’s assessment, followed by the inevitable reference to Haiman (1980), is a clear understatement of the facts: Cognitive Linguistics (which is the “minority view” he refers to) is no longer a minority view.[5] Goddard goes on to say that “theorists like Charles Fillmore and Ronald Langacker believe that knowledge of all kinds is integrated in the mind to such an extent that it doesn’t make any sense to partition it into two distinct realms”. Langacker, as is well known, is one of the founding fathers of Cognitive Linguistics as it exists today – together with George Lakoff, Gilles Fauconnier and Len Talmy. Fillmore is now also very much a part of the Cognitive Linguistics enterprise, even though Lakoff, for instance, has his doubts (cf. Pires de Oliveira forthcoming).

Few have put the Cognitive Linguistics position as emphatically as Langacker (1987:154-166). He claimed – and still does – that the distinction between semantics and pragmatics (which he equates with that between linguistic and extralinguistic knowledge) is “largely artifactual”: there is only one “viable conception” of linguistic semantics, namely the one that a) avoids “false dichotomies” such as those just referred to, and b) adopts an unashamedly encyclopedic perspective. The second feature follows from the first. According to Allan (1995:294), concepts such as Fauconnier’s “mental spaces” (e.g. Fauconnier 1994), Lakoff’s “idealised cognitive models” (or ICM’s; e.g. Lakoff 1987), and Fillmore’s “frames” (e.g. Fillmore 1982, 1985) “are by no means all identical, but they call extensively upon encyclopedic knowledge” (cf. also Allan 1992:357).[6]

According to Geeraerts (1988a), the need to abandon the distinction between lexicon and encyclopedia is a direct result of the necessity to study lexical concepts as an integral part of human cognition in general, instead of as part of an autonomous language structure within human cognition. This goes against the structuralist hypothesis of a lexicon containing a strictly semantic structure that exists independently of the general cognitive organisation of the human mind. The structuralist belief that it was the task of semantics to reveal that independent semantic structure implies that semantics itself is an autonomous discipline. This is not at all what Cognitive Linguistics (and Cognitive Semantics in particular) takes to be the case: its position is that the only appropriate way to study lexical concepts is with constant reference to the general cognitive abilities of humans. There is no separate, specifically linguistic or semantic organisation of knowledge. Research in lexical semantics can therefore not be carried out in an autonomous fashion, or in isolation. It must be carried out in strict cooperation with other disciplines with an interest in the human mind (such as psychology, artificial intelligence, neuropsychology and cultural anthropology).

Geeraerts (1988a) puts the distinction between lexical and encyclopedic knowledge on a par with that between essence and accidence, or that between analytic and synthetic. Both of these, and a few more, had also been cited by Haiman (1980), in his demonstration of the theoretical impossibility of a distinction between dictionaries and encyclopedias (cf. section 3). Haiman is not a lexicographer, but Geeraerts is. In fact, he was one even before he closed ranks with Lakoff and Langacker, and before he became the founding editor of the journal Cognitive Linguistics. It is interesting to note that, as a lexicographer, he published at least one paper (Geeraerts 1985) in which he happily juxtaposed dictionaries, encyclopedias, and encyclopedic dictionaries, and saw no reason to abandon the distinction. I do not believe that Geeraerts would feel his more recent theoretical stance to be contrary to his earlier work. Although he does not appear to have said so anywhere in print, it seems that his position is not unlike that of Haiman (1980), who is mentioned in Geeraerts (1988a) but not in Geeraerts (1985). In other words, there is no theoretical justification for a distinction, but in practice it does not work out that way.[7]

2. Lakoff and Fillmore… before and after

Four years before Lakoff & Thompson (1975) first used the term cognitive linguistics to refer to a programme which remains very much that of Cognitive Linguistics today (cf. Peeters forthcoming a for details), Lakoff (1971:329), at the time a generative semanticist, claimed that Chomsky’s original notion of “strict grammaticality (or degrees thereof)”, which he applied to sentences in isolation, had to be supplemented with a notion of “relative well-formedness”, applied to utterances produced in a context, utterances said to be well-formed relative to “certain presuppositions about the nature of the world”.[8] Lakoff declared *The salami is sleeping ill-formed with respect to our “knowledge of the world”, which is one in which salamis do not sleep; *I went Boston to, on the other hand, involved a breach of the rules of grammar, and therefore was a case of straight ungrammaticality. He then went on to say (ibid.):

It should be pointed out at the outset that [my] claim does not constitute a position that linguistic knowledge cannot be separated from knowledge of the world. On the contrary, it is a claim that the general principles by which a speaker pairs a sentence with those presuppositions required for it to be well-formed are part of his linguistic knowledge.

This almost sounds like Lakoff talking to us in a previous life. Cognitive Linguistics is not too far away, though – both historically speaking (cf. above) and from an ideological point of view. Indeed, Lakoff’s assumption, as Keesing (1979:16) very aptly remarked just over twenty years ago, was that “information about cultural and sociolinguistic conventions required to interpret grammatical patterns [could] be incorporated by stretching formal grammatical theory” (emphasis added). Lakoff was not yet ready to “break down the boundary between a speaker’s ‘knowledge of the language’ and his/her ‘knowledge of the world’”, but he was starting to widen the boundary of the former (Keesing 1979:17). In an interview published in 1974, Lakoff clearly marked his awareness of what had been happening:

We have found that one cannot just set up artificial boundaries and rule out of the study of language such things as human reasoning, context, social interaction, deixis, fuzziness, sarcasm, discourse types, fragments, variation among speakers, etc. Each time we have set up an artificial boundary, we have found some phenomenon that shows that it has to be removed. That is not to say that there are no bounds on the study of linguistics. I only suggest that at this point in history the boundaries are disappearing daily, and one should not be too surprised if the domain of the field continues to expand. (Lakoff 1974:178)

Cognitive Linguistics was basically the natural outcome of this ongoing process: boundaries kept disappearing – at least in the eyes of Lakoff and those who worked alongside him – until it was realised that there were none left. Of course, there were many who would have no bar of it. Hence, Lakoff’s lengthy criticism, in what for a linguistics text was a remarkable best-seller with a remarkable title (Women, fire, and dangerous things, Lakoff 1987), of what he referred (and still refers) to as the “objectivist paradigm in linguistics”, a paradigm he once belonged to, but as a Cognitive Linguist no longer does. Lakoff’s current view remains identical to the one he defended in 1987, when he pointed out that the “dictionary-encyclopedia distinction”, or the objectivist linguists’ distinction between definitional knowledge of words (i.e. knowledge that “corresponds to the essential properties of the entities and categories that the words designate”) and encyclopedic knowledge of words (i.e. knowledge that “corresponds to the contingent properties of the entities and properties [sic] that the words designate”), is a “technical distinction, induced by the rest of the objectivist paradigm”, a consequence of “the objectivist paradigm extended to include language as a matter of objective institutional fact” (Lakoff 1987:172). Elsewhere in the same volume, Lakoff (1987:138) confirmed his previous affiliation with the objectivist paradigm by referring to an earlier paper (Lakoff 1972) where he had made a distinction between definitional properties and characteristic but incidental properties. “This – he added – corresponds to the semantics-pragmatics distinction in the objectivist paradigm, the distinction between what the word ‘really means’ and encyclopedic knowledge that you happen to have about the things the word refers to”.

In the case of Fillmore, it is a similar story. He used the terms dictionary and encyclopedia from the very beginning. The former was supposed to contain “lexical information about words”, the latter “non-lexical information about things” (Fillmore 1969:124). The need to distinguish the two kinds of information was taken for granted, and illustrated by means of two examples, one of which will be left for later (section 5). The other example was based on observations made by Bierwisch (1967), whose name became eventually associated with what has been called “the two-level model”, in which lexicon and encyclopedia are rigorously kept apart (cf. Taylor, this volume; Born-Rauchenecker, this volume; see also my comments on Taylor in section 9 below). Fillmore’s observations deserve to be quoted at length.

Let us examine some of the ways in which users of English speak of the horizontal dimensions of pieces of furniture. If we consider a sofa, a table, and a chest of drawers, we note first of all that a sofa or a chest of drawers has one vertical face that can be appropriately called its front, but the table does not. For a non-vertically-oriented oblong object that does not have a natural front, its shorter dimension is spoken of as its width, the longer dimension as its length. For the two items that do have a front, the dimension along that front is the width (even though it may be the longer of the two dimensions), the dimension perpendicular to the front is its depth.

Objects with fronts, furthermore, are typically conceived of as confronted from the outside, as is the case with the chest of drawers, or as viewed from the inside, as with the sofa. The terms left and right are used according to this inner or outer orientation. Thus the left drawer of a chest of drawers is what would be to our left as we faced it, the left arm of a sofa is what would be to our right as we face it.

This information is clearly related to facts about the objects themselves and the ways in which they are treated in our culture, and cannot be something that needs to be stated as lexically specific information about the nouns that name them. It seems to me, therefore, that the truly lexical information suggested by these examples is the information that must be assigned to the words left, right, wide, long and deep (and their derivatives), and that the facts just reviewed about the items of furniture are facts about how these objects are treated by members of our culture and are therefore proper to an encyclopedia rather than a dictionary. (Fillmore 1969:124-125)

Throughout the seventies, Fillmore remained convinced of the need for a distinction between linguistic and encyclopedic information. He raised the topic in Fillmore (1977a), using the terms that I have just indicated, as well as those he had used in 1969 (dictionary and encyclopedia). Unlike before, he also turned his attention to real dictionaries and real encyclopedias (Fillmore 1977a:132-133):

A frequent topic of discussion among semanticists is the issue of where and how to draw the line between linguistic information about the meanings of words and real-world information about the properties of things. This issue usually takes the form of the question “What is the difference between a dictionary and an encyclopedia?” The famous Spanish Academy definition of dog, as the species in which the male urinates by raising one leg, or the common dictionary definitions of left and right, which speak of the side of a person facing south or north when that person is facing west, clearly are not conceptual analyses of their definienda, but rather serve as recognition tests for people who need to make sure what kinds of things the words designate. It is frequently assumed by linguistic semanticists that the linguist’s job is to determine the purely linguistic information about word meanings, and that a distinction between a dictionary and an encyclopedia can in principle be established. A more realistic view might be something like this: there are things in the world, there are typical event types that one can observe in the world, and there are institutions and cultural values that make human endeavors interpretable; for a very large part of the vocabulary of our languages, the only form a definition can take is that of pointing to these things and actions and institutions and indicating the words used for naming and describing parts and aspects of them.

Fillmore did not conclude that the job of a lexicographer cannot be distinguished from that of an encyclopedist; in fact, as I foreshadowed in my introduction to Fillmore (1977a) a moment ago, he emphasised that, notwithstanding the blurred nature of the distinction between dictionaries and encyclopedias in daily lexicographical practice, the distinction between linguistic and encyclopedic information is a necessary one. Haiman (1980), three years later, would reach a dramatically different conclusion (cf. sections 1 and 3): namely, that the distinction between dictionaries and encyclopedias works very well in practice, even though there is no theoretical basis for a distinction between lexical and encyclopedic knowledge. Fillmore, however, thought that the latter distinction was necessary first of all to provide a convincing account of the interpretative process:

It seems clear, at any rate, that any attempt to relate a person’s knowledge of word meanings to a person’s abilities to interpret texts will have to recognize the importance of nonlinguistic information in the interpretation process. We get clearly different interpretations from the sentences “The fly was on the wall” and “The cat was on the wall”, just because we know different possibilities for stable positions for these two kinds of animals and because we know that the same word – wall – can be used to refer to a vertical surface of a room or building and to a high-relief boundary around a place. Generally this kind of disambiguation is thought of as a use of semantic competence, but in this case it surely involves information of the kind that cannot be sensibly incorporated in the definitions of the associated words. (Fillmore 1977a:133)

The distinction was also necessary to make sense of metaphorical statements (ibid.):

If we hear something like “Harry is a pimple on the face of the community”, we do not use peculiarly linguistic information to interpret what has been said. We know enough about people, pimples, and communities to know that a coherent scene cannot be constructed out of that sentence taken literally. From that detection of a mismatch, we know that we should make use of the psychocultural information that people are embarrassed by and want to get rid of a pimple, and we assume that the speaker intended us to believe that the members of the community have feelings like that about Harry.

A quick look at the extensive Cognitive Linguistics literature on prepositions and on metaphor is enough to indicate that, in the seventies, Fillmore was anything but a Cognitive Linguist in waiting. He did end up joining them, though. To understand how that happened we must look at how frame semantics evolved from an enterprise involving frames and scenes (cf. in the quote above the clause “a coherent scene cannot be constructed”) to one involving (possibly) frames, scenes, schemas and models, and finally to one in which no mention is made of scenes and schemas, because nothing else matters but (redefined) frames.

In at least four papers published in the seventies (Fillmore 1975, 1976, 1977a, 1977b), frames coexisted with scenes. The following passage (Fillmore 1977b:63) shows the difference, but also indicates the author’s dissatisfaction with the latter of the two terms:

I want to say that people, in learning a language, come to associate certain scenes with certain linguistic frames. I intend to use the word scene – a word I am not completely happy with – in a maximally general sense, to include not only visual scenes, but familiar kinds of interpersonal transactions, standard scenarios, familiar layouts, institutional structures, enactive experiences, body image; and, in general, any kind of coherent segment, large or small, of human beliefs, actions, experiences, or imaginings. I intend to use the word frame for referring to any system of linguistic choices – the easiest cases being collections of words, but also including choices of grammatical rules or grammatical categories – that can get associated with prototypical instances of scenes.

Scenes and frames activated each other: upon activation by a word belonging to a certain frame, the corresponding scene would help the addressee activate the word’s meaning. Frames were also associated with other frames, and scenes with other scenes: in the former case, the association would come about “by virtue of shared linguistic material”; in the latter case, it would come about “by virtue of sameness or similarity of the entities or relations or substances in them or their contexts of occurrence” (ibid.).

The notion of “scene” was further expanded in Fillmore (1977a), which is a more recent text than Fillmore (1977b). Towards the end of that more recent text, the author finally admitted that the term scene had been given “too much work to do”. Additional distinctions were necessary (Fillmore 1977a:126-127):

In the first place, we need to recognize the real-world scenes in terms of which people have learned categories and distinctions, and in terms of which people have acquired their original awareness of the objects and experiences that the world has to offer, as well as the real-world scenes that are the contexts and causes of ongoing perceptions and behavior.

Secondly, there are memories and distillations of real-world scenes in people’s minds, possibly restructured in ways provided by their participation in a particular community, possibly with some aspects of them forgotten or suppressed and others enhanced.

Thirdly, there are schemata of concepts, stereotypes of familiar objects and acts, and standard scenarios for familiar actions and events that can be spoken of independently of given individuals’ memories of experiences.

Fourthly, there is the imagined scene of the speaker as he is formulating his text; and fifthly there is the imagined scene of the interpreter as he is trying to construct a model of the world that matches the text he is interpreting.

And lastly there are the sets of linguistic choices that a given language provides and the ways in which these activate or are activated by particular conceptual schemata.

To identify each of these, Fillmore ended up borrowing some of the terms that had been proposed in artificial intelligence.[9] How exactly he proposed to define scenes, schemata, frames, and models is not all that important, however (see Fillmore 1977a:127 for details), because he soon realised that the best way forward was to make no distinctions at all. The only term left in the eighties and beyond (e.g. Fillmore 1982, 1985; Fillmore & Atkins 1992) was frame. In Fillmore’s current thinking, frames provide the conceptual underpinning, the underlying conceptual structure required for all forms of interpretation. Frames, which have become very much part of the conceptual apparatus of Cognitive Linguistics, are comprehensive but stereotypical knowledge stores where the distinction between lexical and encyclopedic knowledge is no longer made. Fillmore has come to share with just about every other Cognitive Linguist under the sun the view referred to by Langacker (1997) as the “doctrine of encyclopedic semantics”.[10]

3. Haiman, Frawley, and Wierzbicka

For the careful onlooker, it should be clear that some Cognitive Linguists have refused to get “indoctrinated”. One of them is Anna Wierzbicka, who is among the most original thinkers in semantics and pragmatics of the last thirty years. Her relevant work is reviewed in what follows, together with that of John Haiman and William Frawley.

Cognitive Linguists typically appear to assume that the first author to have argued in favour of an overhaul of the distinction between lexical and encyclopedic knowledge is Haiman (1980). In the present volume, Joe Hilferty, who recently co-authored a Spanish-language introduction to Cognitive Linguistics (Cuenca & Hilferty 1999), is quite explicit about it. As we shall see, the reality is that Haiman was by no means the first, even though his often-quoted study on dictionaries and encyclopedias is probably the first scholarly paper that was entirely devoted to the topic. In it, he asked what exactly should be mentioned in a dictionary (what is linguistic or lexical knowledge), and what ought to be reserved for the encyclopedia (what is world or encyclopedic knowledge). The view that “knowledge of (the semantics of) a language – properly codified in a dictionary – is distinct from that knowledge of the real world which belongs in an encyclopedia alone” was referred to as an “article of faith” held by many linguists and philosophers, an “apparently uncontroversial distinction” that, in spite of earlier attacks, was “still very far from being a dead horse” (Haiman 1980:330).[11]

Ironically, it was Wierzbicka’s (1972:54) definition of the word horse (!) as “an animal called ‘horse’”, among other similar definitions of natural kind terms provided by the same author and by several others, that had incited Haiman to investigate the matter in more detail than had ever been done before. Haiman found himself in agreement with the thesis that it was impossible to add anything to the definition without turning it from a dictionary definition into an encyclopedia definition. In fact, he went one step further, claiming that “perhaps the linguistically most impeccable definition of elephants, as of all other words, is simply the most austere: elephant NP”, and that all the rest “is contingent on knowledge of the world” (Haiman 1980:342). Of course, dictionaries do add a lot more – their definitions typically only start where according to Haiman they should end – which is why, in the author’s view, dictionaries are in fact encyclopedias (Haiman 1980:331). The conclusion ran as follows (Haiman 1980:355):

Having demonstrated that dictionaries are not in principle different from encyclopedias, I do not, in my wildest dreams, expect that sales and production of either one or the other will come to an end. Part of the reason for this is that the distinction between dictionaries and encyclopedias, while theoretically untenable, has the happy property of working very well in practice.

It was perhaps unavoidable that this conclusion would be questioned in the very pages of the journal that had accepted Haiman’s paper. Frawley (1981) published a rejoinder “in defense of the dictionary”, to which Haiman (1982) issued a short (and, one presumes, rather hastily written) reply. Haiman blamed Frawley for having missed the point he had been trying to make. I am not absolutely convinced. What Frawley did was show that Haiman’s argumentation had been basically flawed: there was nothing wrong with the various distinctions that either had been – or looked as though they might be – invoked to motivate the distinction between linguistic and encyclopedic knowledge, and that Haiman had said were unsustainable. There were six such distinctions:

· sense vs. denotation (also referred to by Haiman 1980 as the distinction between linguistic knowledge and cultural knowledge, no doubt under the influence of Keesing 1979; I am not sure whether the two distinctions are identical);[12]

· subjective vs. objective fact (a distinction apparently upheld in Wierzbicka’s early work, especially in her claim that natural kind terms – because they correspond to objective categories in the world – cannot be linguistically defined);

· essence vs. accidence (or core knowledge vs. peripheral knowledge, à la Bierwisch; not to be confused with core and periphery in prototype semantics – cf. Haiman’s 1982 rebuff of Frawley 1981; cf. also the afterword to Taylor, this volume);[13]

· semantics vs. pragmatics (a distinction abandoned by Cognitive Linguists, as shown in section 1);

· analytic vs. synthetic statements (this distinction will be revisited in section 7);

· proper names vs. common names or definite descriptions (on “what names tell about the lexicon and the encyclopedia”, see Allan 1995, and also this volume).

Haiman’s thesis that the traditional distinction between lexical and encyclopedic knowledge had been motivated by one or more of these allegedly unsustainable distinctions led him to the conclusion that it, too, was unsustainable – and, with it, at least from a theoretical point of view, the distinction between dictionaries and encyclopedias. What he failed to see, and what Frawley did not observe either, is that there were perhaps other ways of justifying the distinction. Wierzbicka eventually confirmed that this was indeed the case. It is probably quite correct to say that Haiman’s paper convinced the Polish-born Australian linguist that her earlier claim regarding natural kind terms (cf. Wierzbicka 1972, 1980) had been misguided. She set out to show that natural kind terms do have a fair amount of linguistic knowledge associated with them, that therefore they can be linguistically defined in a far more satisfactory way than she had assumed before, and that it is entirely legitimate to oppose linguistic and encyclopedic knowledge.

Wierzbicka’s views set her apart from her colleagues in the Cognitive Linguistics movement. She condemns the claim that it is “impossible to draw a line between ‘meaning’ and ‘knowledge’ or between ‘dictionaries’ and ‘encyclopaedias’” for the “unfortunate effect” that it has had on the study of the lexicon (Wierzbicka 1996:336). Similarly, the “belief that a dictionary definition represents nothing other than a selection from a (real or imaginary) encyclopaedia entry, with the choice being determined by practical considerations and having no theoretical justification, leads to stagnation in lexical semantics” (ibid.). Remarks such as these have not prevented Wierzbicka from being considered by many Cognitive Linguists as one of them. Others are noncommittal or question her belonging. In other words, it is not entirely clear whether she is best described as “friend, foe, or fellow traveller” (Goddard forthcoming).

Perhaps, it is fair to consider Wierzbicka a co-opted member rather than a founding member of the Cognitive Linguistics movement, even though she was in Duisburg in the spring of 1989 for a symposium (organised by René Dirven) that was later said to have “marked the birth of cognitive linguistics as a broadly grounded, self-conscious intellectual movement” (Langacker 1990:1). It was at that symposium that, in Langacker’s words, “initiation of the journal Cognitive Linguistics was announced”, that “plans were made to form the International Cognitive Linguistics Association”, and that “agreement was reached to launch the monograph series Cognitive Linguistics Research” (ibid.). Wierzbicka’s contribution to the Duisburg symposium immediately follows Brygida Rudzka-Ostyn’s introduction to the third volume in that series (Geiger & Rudzka-Ostyn 1993). A very thought-provoking paper of hers (now revised and expanded as chapter 10 of Wierzbicka 1996) appears in the very first issue of the journal (Wierzbicka 1990).

Langacker (1990:2) judges Wierzbicka’s work to be “entirely compatible” with his own. Lakoff (1990:45-46), on the other hand, just pages before Wierzbicka’s own article referred to above, spends considerable time explaining how much he disagrees with her (cf. also Lakoff 1987:278-280), how much he learnt from her, and how important it is for all of us to have an ongoing dialogue with people whose work we can partly, but not totally, endorse. Interestingly, in two recent interviews (Pires de Oliveira 1998, forthcoming), Lakoff’s tone is quite different. He does no longer consider Wierzbicka a Cognitive Linguist. In the first interview, she is described as an “idealist”, whereas Cognitive Linguistics is about “the embodiment of meaning”, about “the lack of separation between mind and body”. In the second interview, Lakoff categorically states that Wierzbicka “does not look empirically at the same range of data that cognitive linguists do”. He adds that “her analyses sometimes capture some aspects of meaning, but they miss an awful lot”, and that “they do not fit with what we know of the mind and the brain”. At the same time, in the same interview, talking about the building blocks for non-universal concepts and metaphor systems, he takes on a clearly Wierzbickian perspective – but without saying so.[14]

Wierzbicka has come in for a good deal of criticism from at least one other Cognitive Linguist of the early days, viz. Dirk Geeraerts (cf. section 1). In his review of Wierzbicka (1985), Geeraerts (1988b) does show a lot of restraint; he is slightly more critical in Geeraerts et al. (1994), and outspokenly negative in his recent pastiche of Ancient Greek philosophical discourse (the dialogue between teacher and pupil; cf. Geeraerts 2000).[15] The Belgian linguist is less dismissive, though, than his American counterpart. He does not bar Wierzbicka from the Cognitive Linguistics fold, but opposes within it two methodological extremes, summarised as follows by Goddard (forthcoming):

In the “good corner” there are the data-driven, empirically-minded linguists doing psycholinguistics, neurophysiological modelling and corpus analysis. In the “bad corner” there is the “idealistic tendency” represented by Wierzbicka and her colleagues, with their appeals to intuition and platonistic views about universal conceptual primes.

The data-driven, empirically-minded linguists reject the idea of a distinction between lexicon and encyclopedia, whereas the “idealistic tendency” subscribes to the opposite view. Two publications of Wierzbicka’s deserve to be singled out for further comment in this regard. In her assessment of Wierzbicka (1985), Lehrer (1988:236) writes that Wierzbicka “draws a distinction between a dictionary (what a proper definition contains) and an encyclopedia (additional information about the world)”. She then adds that where Wierzbicka “draws that line is considerably different from that of other lexicologists”. I for one have the feeling that this has not been sufficiently recognised. Born-Rauchenecker (this volume), for instance, points out that Wierzbicka separates the two domains of knowledge, but that – unlike some others – she describes encyclopedic knowledge as far as possible. The so-called encyclopedic knowledge that Wierzbicka describes is encyclopedic knowledge as understood by Born-Rauchenecker, not by Wierzbicka herself. Allan’s (1992:356) observation (repeated in Allan 1995:293) according to which “Wierzbicka (1985) developed semantic descriptions very reminiscent of those in an encyclopedia” (emphasis added) is more accurate than Born-Rauchenecker’s statement: while Wierzbicka claims to be describing linguistic knowledge, the fact remains that some of the detail she provides has a distinctive encyclopedic flavour (Lehrer 1988:236; cf. also the afterword to Taylor, this volume).[16]

The 1985 book contains a lot of rather lengthy lexicographical definitions, mostly of natural kind terms and of cultural kind terms, far too long for the average dictionary, but nonetheless consisting exclusively of what Wierzbicka takes to be linguistic information. In her view, an important difference between linguistic and encyclopedic knowledge (the latter term is used quite sparingly) is that “linguistic knowledge is essentially shared between all the speakers of a language, whereas real-world knowledge is not” (Goddard 1998:14). This leaves of course a lot of uncertainty regarding so-called “common knowledge”, e.g. “that dogs have four legs, bark, and wag their tails” (ibid.). What sort of knowledge is this? Wierzbicka’s answer is that the meaning of the word dog (to take but one example) coincides with the folk knowledge surrounding dogs. This folk knowledge includes among other things the features quoted by Goddard. “The linguistic concept of dog, for instance, includes barking, tail-wagging, and much more besides” (Goddard 1998:15).

Wierzbicka’s position is perhaps more clearly stated in her 1995 paper (revised reprint in Wierzbicka 1996), where terms such as encyclopedia and encyclopedic knowledge are no longer rare occurrences:

I am using the words “dictionary” and “encyclopaedia” in a metaphorical sense, referring to language-related “folk knowledge” (everyday knowledge) and to language-independent scientific knowledge (and certainly not to any concrete reference works such as the Oxford English Dictionary or the Encyclopaedia Britannica). (Wierzbicka 1996:337)

According to Wierzbicka, there is linguistic evidence that the human mind itself draws a distinction between a “mental dictionary” and a “mental encyclopedia”.[17] The nature of the available linguistic evidence is neatly summarised in Taylor (forthcoming). He argues that “acquisition is not a process of building up a concept from its constituent parts”, but that it consists in the “gradual elaboration of a knowledge network”. He then goes on as follows:

Wierzbicka’s definition of mouse is instructive in this respect. The definition, as mentioned, extends over almost two pages, and includes such information as the characteristic size, shape, and colour of mice, their habitat, their manner of moving, and the sounds they make. Also included is the fact that mice are (or are thought of as being) timid, quiet, and inconspicuous; that cats chase them; that they are fond of eating cheese; that they live near humans; that they are regarded as pests; and that people try to get rid of them. (Surprisingly, the fact that some people – stereotypically women – have a phobic terror of mice, is not included in the definition.) Wierzbicka motivates the contents of her definition largely on linguistic grounds. A cat can be ‘a good mouser’; poor quality Cheddar cheese can be called (or used to be called) ‘mousetrap cheese’; a shy, timid, and inconspicuous person (usually female) can be called ‘a mouse’ (or ‘a grey mouse’); and so on. Idioms (‘as poor as a church mouse’) and nursery rhymes (‘Three blind mice’) are also called in evidence. Excluded from the definition are ‘encyclopedic’ facts about mice that are not reflected in everyday linguistic usage, such as their geographical distribution, the length of the gestation period of the female mouse, the size of the mouse litter, and such like.

Wierzbicka’s linguistic definition of mice, first attempted in Wierzbicka (1985:175-176), later revised in Wierzbicka (1995, 1996:340-341), “sums over a range of conventionalised uses of the word mouse; these include idioms, fixed expressions, typical collocations, standard metaphorical uses, and so on” (Taylor forthcoming).[18] The differences between the 1985 and 1995/1996 definitions are quite remarkable, and would in themselves be worthy of detailed study. Why, for instance, does the 1985 version (which Taylor clearly did not consult) refer to the fact that “people say that women are often frightened of contact with them” (Wierzbicka 1985:176), whereas no mention is made of this in the later definition?

4. Raskin

Wierzbicka’s view that lexical and encyclopedic knowledge should be kept apart is shared by Victor Raskin who, apart from being a contributor to the present volume, is the author of a paper on linguistic and encyclopedic knowledge in text processing (Raskin 1985a).[19] Major topics in the area of text processing and in that of text linguistics at large are cohesion and coherence (cf. Peeters 1994); although there is a lot of confusion in the literature as to what each of these concepts covers, the fact remains that they are necessary ingredients of a text (and of course also of single utterances).[20] At times, cohesion and/or coherence result from linguistic knowledge, as in (1), at other times they result from encyclopedic knowledge, as in (2). Both (1) and (2) are taken from Raskin (1985a:93).

(1) I did not like the spoon. The bowl was too small.

(2) I did not like the spoon because of the ornament.

How do we know that native speakers use linguistic knowledge to attribute coherence to (1), and encyclopedic knowledge to do the same in the case of (2)? Raskin’s answer is based on what one typically finds in a lexicographical definition. Since dictionary entries are intended by lexicographers to represent the meaning of words, and since a typical dictionary entry for the word spoon refers to an ‘eating or cooking implement consisting of a small shallow bowl with a handle’, without adding anything on the ornaments that one sometimes finds on spoons, the attribution of coherence to (1) involves linguistic knowledge, whereas the attribution of coherence to (2) involves encyclopedic knowledge. Linguistic knowledge is defined as knowledge “internalized by the native speaker of a language by virtue of his/her knowing the language in question”; it includes “familiarity with the meanings of the words and of the ways the words can be combined together”. Encyclopedic knowledge, on the other hand, is defined as “what the native speaker knows about the world he/she lives in and what is not included in his/her linguistic knowledge” (Raskin 1985a:92).

Raskin lists a number of arguments against the distinction followed by a number of arguments in favour of it. Since references are not always provided, it is not clear whether all or only a selection were gleaned from the available literature. Raskin (1985a:99) does find, though, that most of them “are not easily defensible”. In what follows, I only reproduce the more interesting ones, with the additional exclusion of arguments from within artificial intelligence, an area to which Raskin has made important contributions (see section 9.5 below).

Because of the widespread belief that frame structures (cf. section 1 and footnote 5) provide a convenient format (perhaps the most convenient format) for the representation of both linguistic and encyclopedic knowledge, it has been argued (e.g. by Hudson 1985) that no distinction should be made. The opposite belief, namely that encyclopedic knowledge cannot be as easily represented by means of frames as linguistic knowledge (for instance because frames are too rigid or too formal), may lead to the view that a distinction is after all in order. The two arguments do not only cancel one another out; according to Raskin, they are faulty. Identical representation does not necessarily preclude differences along other dimensions, and divergent representations do not necessarily rule out the existence of a different format of representation that suits both kinds of knowledge to perfection, and that therefore makes their identity obvious.[21]

· Lexicography requires a distinction between linguistic and encyclopedic knowledge for practical reasons: dictionary entries must be finite and reasonably succinct. The lexicographer’s task is not feasible, unless a limit is placed on the amount and the sort of information that is allowed to appear in a dictionary. Linguists such as Eva Born-Rauchenecker and Anna Wierzbicka (cf. section 3, as well as Born-Rauchenecker, this volume) also argue that, in lexicography, the distinction is an important one. Raskin (1985a:99), however, rejects the lexicographer’s demands. Lexicography, he says, is “anecdotal, circular, and devoid of any scholarly value”; it has ignored semantics for centuries, and has no right to demand anything now.

· Variations from one subject to another (a natural consequence of the inevitably variable size of their respective vocabularies) appear to make it impossible to decide once and for all which pieces of knowledge are linguistic. Moreover, it seems rather unlikely to suggest that, when a new word is acquired by a speaker, particular pieces of knowledge that were already available to that speaker would change their status (i.e. move from encyclopedic to linguistic knowledge). According to Raskin (1985a:100), this argument must be rejected on the grounds that a similar one would have to lead to the denial of the existence of English (or of course any other language). No two people speak in exactly the same way, yet to think in terms of the language of a certain speech community, irrespective of any idiosyncracies and individual variations, is far more than a useful fiction.

The reason why, in Raskin’s opinion, “theoretical linguistics should be concerned about the distinction in question while the applied fields [especially lexicography and artificial intelligence, B.P.] can continue feigning nonchalance about it” (Raskin 1985a:100) is not covered in the above list of arguments: basically, the amount of semantic information allowed to take part in any formal procedure of semantic interpretation associated with a formal semantic theory has to be limited “simply because such a theory is likely to be a mechanical symbol-manipulation device and such a device cannot manipulate infinite sequences of symbols” (ibid.). Raskin proposes a theory “which, in principle, can accommodate and represent both linguistic and encyclopedic information but which can also distinguish between them and only use the former for semantic calculation” (Raskin 1985a:101). At the same time, he strongly believes that the applied fields should in fact not be too dismissive of what is happening in linguistic theory. More than many others, Raskin is aware of the fact that theoretical distinctions that are of no practical use should not be made in the first place. The fact that he makes the distinction strongly suggests that he believes it ought to be made, for theoretical as well as for practical reasons. In his co-authored contribution to the present volume, he shows that in machine translation a distinction is most definitely required.

5. Pustejovsky, Fillmore (bis), and Klinkenberg

Several contributors to this volume (Reboul, Taylor [afterword], Larrivée, Allan, Murphy) acknowledge the impact that James Pustejovsky’s Generative Lexicon model (most forcefully expounded in Pustejovsky 1995) appears to have had in the last decade. His work has been rightly noted and hotly debated in natural cognition as well as artificial cognition. One of the most stinging and effective critiques of Pustejovsky’s model is arguably contained in Nirenburg & Raskin (1996). According to Reboul (this volume), the model “has done much to revive the controversy around the boundary between lexical and encyclopedic knowledge”. In fact, the boundary he refers to is “between what we formally take to be linguistic or lexical knowledge and that what is sometimes referred to as ‘commonsense knowledge’” (Pustejovsky 1995:232). What exactly is meant by the latter term is not entirely clear. It seems, though, that a lot of encyclopedic knowledge, as usually defined, is not exactly shared by everyone, or even by a majority, and therefore does not qualify as commonsense knowledge. At best, the latter seems to be a subset of the former. Reboul further points out, and quite rightly so, that Pustejovsky’s originality lies in his decision to treat as part of the (generative) lexicon various pieces of knowledge that lexicologists before him used to almost automatically relegate to the encyclopedia.[22]

As a first illustration, let us revisit Fillmore (1969) (for more, see section 2 above). Building on work done by Katz (1964), Fillmore defined evaluative features as those features that describe from what point of view someone or something is evaluated as being good or bad, poor or excellent, etc. He argued that there are several clear-cut instances where “the evaluative feature can be automatically specified from the function-identifying part of a definition” (Fillmore 1969:123). Fillmore cited the case of agentive and instrumental nouns, whose definition refers to one or the other activity. A pilot is someone who navigates an air vessel. This is Fillmore’s gloss, not mine; it is lexical knowledge, it gives us the meaning of the word pilot. By default, a “good pilot” will be someone who is “good at navigating air vessels”. Similarly, a knife is an instrument used to cut things. A “good knife” is a knife that “cuts well”. On the other hand, Fillmore (1969:124) also believed that there are clear-cut instances where “the evaluative feature apparently needs to be specified separately”. Nothing in the lexical definition of the words food or photograph specifies in a direct or immediate way what it means for food or for a photograph to be called good (its nutritional value or palatability in the case of food, its clarity or ability to “elicit positive esthetic responses in the viewer” in the case of a photograph). Fillmore (ibid.) clearly thought that these were important matters, because they raise an important question:

The question a lexicographer must face is whether these matters have to do with what one knows, as a speaker of a language, about the words in that language, or what one knows, as a member of a culture, about the objects, beliefs and practices of that culture. Do we know about books that they are used in our culture to reveal information or elicit certain kinds of esthetic appreciation, or do we know about the word book that it contains evaluative features that allow us to interpret the phrase a good book? Do we understand the expression good water (as water that is safe for drinking) because its semantic description has set aside that one use of water as the use in terms of which water is to be generally evaluated, or because we know that for most purposes (e.g., watering the grass, bathing) any kind of water will do, but for drinking purposes some water is acceptable and some is not? These are serious questions, but we can of course avoid facing them by making, with the typical lexicographer, the decision not to insist on a strict separation between a dictionary and an encyclopedia. (Fillmore 1969:124)

The impression one gets from reading Fillmore is: a) that he was unsure whether addressees rely on lexical or on encyclopedic knowledge to work out the meaning of noun phrases such as a good book, or good water; b) that he thought that interpreting noun phrases such as a good pilot and a good knife involves lexical knowledge; and c) that encyclopedic knowledge is needed to make sense of noun phrases such as good food and a good photograph.

What is Pustejovsky’s view on the matter? Before I answer that question, I would like to point out that, although included in the bibliography, Fillmore (1969) does not appear to be quoted anywhere in Pustejovsky’s text.[23] However, very similar examples are used: a good car, a good meal, a good knife in Pustejovsky (1995:ch.3); a good umbrella, a good meal, a good teacher in Pustejovsky (1995:ch.4). Then, in chapter 7, a generative mechanism called selective binding is introduced, which treats the adjective as a function and applies it to a particular “quale” or semantic role within the noun that it qualifies (cf. Pustejovsky 1995:76-81, or Reboul, this volume, for further details on a noun’s qualia structure). This is followed by the statement that the same interpretive mechanism allows us to account for the contextualised senses of adjectives in general. In other words, the only knowledge that is needed is lexical.

Type coercion is another generative mechanism that operates in the generative lexicon (Pustejovsky 1995:106-122). Like selective binding, it appears to allow for the view that the interpretation of certain phrases, instead of relying on encyclopedic (or rather commonsense) knowledge, involves only lexical knowledge. Type coercion is very similar to what happens, according to Klinkenberg (1983, 1984), when so-called allotopy leads to reassessment in order to restore the isotopy induced by a verb. The crucial difference is that, when allotopy occurs, encyclopedic knowledge is used to reestablish isotopy. The terminology will be explained in a moment; first, I wish to express my surprise at the fact that neither Reboul (this volume), nor various other French-speaking authors who have commented on Pustejovsky (e.g. Danièle Godard, Jacques Jayez, Georges Kleiber; cf. Godard & Jayez 1993a, 1993b, Jayez 1996, Kleiber 1997, 1999) or collaborated with him (e.g. Pierrette Bouillon; cf. Pustejovsky & Bouillon 1995), appear to be aware of Klinkenberg’s work – or of that of the so-called Groupe m to which he belongs. The time has come to make that work better known outside the semiotic-rhetorical circles where it was originally conceived.[24]

One year before Haiman (1980) saw the light of day, Klinkenberg spoke about semantic and encyclopedic knowledge at a Vienna semiotics conference. He argued, unlike Haiman, that the two were distinct. Parts of the conference paper, published five years after the event (Klinkenberg 1984) were used in another relevant text (Klinkenberg 1983), published the year before. I will mainly refer to the latter. Klinkenberg’s (1983:295) demonstration revolves around the French utterance Je viens de relire Greimas ‘I have just been rereading Greimas’. The verb lire ‘read’ carries an isotopy requirement: its direct object must refer to a written item. Since the object perceived by the addressee does not meet the isotopy requirement, it is allotopic and in need of reassessment. The reassessment process produces a so-called conceived degree (un degré conçu), namely ‘written item emanating from Greimas’, which is superposed onto the original so-called perceived degree (degré perçu). Both are part of what may be conventionally called the Greimas universe, which includes items such as ‘writings’ by Greimas, ‘classes’ he teaches, ‘views’ he defends, ‘disciples’ – and of course Greimas as a ‘person’ is in it as well. Commutation of the item ‘person’ with any of the others allows the formation of utterances such as I have reread the whole of Greimas, meaning ‘I have reread all his writings’, I’ve got Greimas this afternoon ‘I am attending one of his classes’, That’s typically Greimas ‘that’s one of the views he defends’ etc.

According to Klinkenberg, the deployment during interpretation of an utterance with an allotopic object reveals that, predating the utterance, there has to be a representation of the world, a representation that implies among other things one’s acquaintance with a person called Greimas, a knowledge of writings, and their attribution to that person. All of these are non-linguistic elements. Klinkenberg hesitates to refer to them as “elements of knowledge”, since he believes that the word knowledge has too much of a passive connotation. He qualifies ‘attributing some piece of writing to a person’ as a highly complex operation that entails a decision on the person, one on the written piece, one on the relationship between the person and the written piece, even one on the attribution (itself a complex operation). The common belonging of the various items to a coherent universe is not at all a given, even less a linguistic given: it is nothing short of a set of decisions that aim to lend coherence to what are basically isolates.

The same sort of remark applies to the articulation or structuration of the preexisting universe. Not all commutations within it are equally felicitous. Greimas being of Lithuanian origin, ‘Lithuania’ is no doubt part of the Greimas universe, but the utterance I’m going to Greimas is not very likely to trigger an interpretation of the type ‘I’m going on a pilgrimage to Greimas’ birthplace’. The various items in a universe are not merely juxtaposed, but organised in intricate ways, according to certain relations of implication, subordination, etc. That organisation, as well as the common belonging referred to in the previous paragraph, are derived from our knowledge of the world: this is, according to Klinkenberg, encyclopedic knowledge situated “en deça du composant sémantique” ‘beneath the semantic component’, and to be embedded in the linguistic description (because it is part of the speaker’s competence; cf. section 9.2). Technically, the “encyclopedic component” stores knowledge “accepted” by the language community, and therefore always determined by social and historical factors. This, in turn, explains a major originality of Klinkenberg’s encyclopedia, namely the fact that it explicitly allows for the storage of contradictory pieces of knowledge. There are two opposing perspectives: a practical one which is mostly linked with immediate perception, and a theoretical one where the link with perception is more tenuous. When we talk about a fire, we often visualise and/or conceptualise it together with the smoke that escapes from it: in practical terms, the smoke is “part of the fire”, together with the flames, the heat, the combustion etc. On the other hand, it is equally possible, though perhaps not so obvious, to keep the fire and the smoke separate. For more details, see Klinkenberg (1983:296-297).[25]

Pustejovsky’s examples of type coercion are not identical to Klinkenberg’s, but they are close. True to the generative tradition of linguistics, Pustejovsky distinguishes between a deep and a surface level of syntax. At the deep level, predicates assign semantic types to their arguments. The verb begin, for instance, assigns the type ‘event’. Type coercion explains why it is nonetheless possible, at surface level, to have constructions of the type “NP1 begin NP2”, where the latter NP appears to be of the wrong type (‘object’ instead of ‘event’). The mechanism is entirely controlled by the predicate. It does not apply when begin is followed by an infinitive (as in John began to read the book), because the infinitive already refers to an event. When there is no infinitive, and begin is followed by a noun complement, type coercion may apply. It does only if the NP is of the object type; in that case, another type (viz. ‘event’) is superimposed upon it. In other words, the NP is coerced to refer to an event rather than to an object. Obviously, if the NP is of the correct type (as is the case in, e.g., begin a class (said of a teacher; a class is an event), no type coercion is required.

The question is: when a new type is generated, how exactly does this happen? In Pustejovsky’s model, type coercion critically relies on lexical information associated with the noun complement. Again, as previously illustrated with reference to selective binding, what is exploited is the so-called qualia structure of the noun. In the case of a noun such as book, there is a telic quale identifying what books are for (they are meant to be read); there is also an agentive quale identifying who is at the origin of the object book (a writer). Depending on the context, either the verb read or the verb write are “drawn” from the qualia structure associated with the noun book, and their type (which is the expected one after a verb such as begin) is superposed upon the type of the NP book. For relevant criticism, cf. Reboul (this volume), as well as the various “French sources” mentioned above (except, of course, Pustejovsky & Bouillon 1995).

6. Distributed Morphology

Whereas Pustejovsky stores in the lexicon information that many would rather treat as encyclopedic, those behind another fairly new model have done the exact opposite. The model, which goes back to the early 1990s, is known as Distributed Morphology, and is basically the brainchild of Morris Halle and Alec Marantz. No single authoritative book-length presentation of its basic principles is as yet in existence, even though there is a substantive body of research that has already been carried out.[26] A crucial tenet of the theory is that there is no lexicon as understood by most if not all forms of generative grammar practiced in the 1970s and beyond. In the absence of a unique lexicon, there are also no lexical items, and there is no lexical knowledge. Marantz (1997:203) explains that “Distributed Morphology explodes the Lexicon and includes a number of distributed, non-computational lists as Lexicon-replacements”. Harley & Noyer (forthcoming) point out that “the jobs assigned to the Lexicon component in earlier theories are distributed through various other components”. Statements such as these go a long way towards explaining how the model arrived at its name.

Distributed Morphology distinguishes two types of meaning (syntactic and encyclopedic). The various abstract morphosyntactic features (e.g. [Det], [1st], [cause], [+ pst], [Root], [pl], etc.) combined by the syntax into so-called syntactic terminals or nodes have no meaning other than syntactic. They constitute what might be called a “narrow lexicon”, which “most directly replaces the Lexicon as it provides the units that the syntax operates with” (Marantz 1997:203). The same features are also part of so-called vocabulary items. These may be identical in size to what is traditionally referred to as words, but also either smaller or larger. Vocabulary items are the building blocks of morphophonology. Here, the role of the features is to indicate where (in which syntactic terminals or nodes) particular items may be inserted. Items of a given category “compete” for insertion into a node belonging to the same category. The vocabulary item that presents the best match against the features of a specific node wins out.

Importantly, vocabulary items consist of nothing more than a phonological string (e.g. /dOg/) combined with feature bundles, and therefore syntactic meaning, of the sort referred to (in the case of /dOg/: [Root], [+ count], [+ animate]). They do not in themselves have any other meaning, but are associated with an encyclopedic entry. Encyclopedic entries provide full details regarding the second type of meaning distinguished within Distributed Morphology, namely encyclopedic meaning. This second type of meaning plays no role whatsoever in the syntax. The encyclopedic entry for the vocabulary item dog specifies, among other things, that dogs are four-legged canine pets that may bite and that enjoy chasing balls. There is a comparable encyclopedic entry for, say, cat – which entails, among other things, that the meaning components that distinguish cat from dog are encyclopedic. Encyclopedic knowledge is basically non-linguistic knowledge, which means that in Distributed Morphology there is no linguistic difference between the items dog and cat.[27] Encyclopedic knowledge also specifies, for instance, that, in the environment let sleeping ___ lie, dog refers to a discourse entity that is better left alone. The precise relationship between what are basically the three Lexicon-replacements – i.e. the encyclopedia, or sum total of all encyclopedic entries, the vocabulary, or set of all vocabulary items, and the narrow lexicon (cf. above) – remains a topic of discussion (cf. Marantz 1997, and also Harley & Noyer, this volume).[28]

7. Katz & Fodor’s legacy

Allan (1992:356 = 1995:293) notes that, before the 1980s, it was “normal practice” to favour “parsimonious dictionary knowledge” against “elaborated encyclopedic knowledge”. This is perhaps a slight overstatement, but it can be seen as a direct reference to Katz & Fodor (1963), and to other related work such as Katz & Postal (1964) and Katz (1966), none of which is referred to explicitly. What Katz and colleagues (esp. Fodor) had to say was important, though, as clearly recognised by Raskin (1985a:97-98):

It would be fair to say that the issue [of linguistic vs. encyclopedic knowledge; B.P.] was brought into contemporary semantic theory by Katz and Fodor [1963] who made a strong and clear claim about the boundary between one’s knowledge of language and one’s knowledge of the world. Not only did they claim the existence and necessity of such a boundary but they also declared that the boundary served another important purpose – it separated the object of semantic theory from all the rest. Katz and Fodor believed that it was impossible for any formal semantic theory to account for the meaning of any sentence in any context because, according to them, that would amount to the formal description of one’s entire knowledge of the world, the goal which they considered axiomatically unattainable. The object of semantic theory for them was the meaning of the sentence in isolation, and what it took to calculate it was linguistic knowledge while the rest was not.

Raskin (1985a:100) endorsed the need for a clear delimitation (cf. section 4), but rejected Katz & Fodor’s reasons, arguing that to narrow the scope of semantic theory to the “more feasible but almost entirely uninteresting task of accounting for the meaning of the sentence in isolation” is absurd. He went on to say that if the goal of semantic theory is “to match the native speaker’s semantic competence”, Katz & Fodor’s proposals had to be “self-defeating”: no native speaker ever deals with sentences in isolation. However, for Katz & Fodor, all that mattered was the dictionary, their name for the semantic subcomponent in the theory responsible for linking linguistic forms and meanings.

As it turns out, the integrity of the model was remarkably short-lived. One of the reasons why it came in for a lot of criticism fairly early in the piece has no doubt to do with what many perceived as the spurious distinction between so-called markers and so-called distinguishers. For Bolinger (1965), who listed a string of problems left unaddressed by Katz & Fodor, it was one of two unjustified dualisms, the other one being that between knowledge of the language and knowledge of the world. Both deserve a closer look: contrary to appearances, both are relevant for the lexicon-encyclopedia debate.

First to markers and distinguishers. Before reiterating what Bolinger had to say about them, I should point out, with Allan (1986:306-314), that between 1963 and 1972 Katz shifted ground on a number of occasions: the definition of what was a marker and what was a distinguisher did not remain unchanged.[29] Allan also notes that the term distinguisher vanished from Katz’s work after 1972. It was in his Semantic theory that, for the last time, Katz sought to clarify what the much-maligned distinction between markers and distinguishers had intended to achieve. No reference at all is made in the relevant pages (Katz 1972:82-88) to encyclopedic information or world knowledge, which strongly suggests that markers and distinguishers were both intended to identify linguistic information. Both were supposed to be semantic features, with a distinguisher being whatever is left over of a given sense within a word once all the markers have been taken care of. Square brackets were used to identify the former, round brackets to identify the latter, as in the following oft-quoted semantic description of the English word bachelor, taken from Katz & Fodor (1963).

bachelor

!

noun

qp

(Human) (Animal)

ep p

(Male) [who has the first or (Male)

qi lowest academic degree] p

[who has (Young) (Young)

never married] ! !

[knight serving under [fur seal when without a

the standard of another mate during the breeding

knight] time]

Returning to Bolinger now, we may note that he used the above bachelor analysis to argue that markers and distinguishers could be shown to be one and the same thing, with all the distinguishers reducible to markers (Bolinger 1965:558-562). Importantly, though, in the course of the argument the author granted the possibility of “a spot for distinguishers, or something like them”, situated “a bit to one side” of where Katz & Fodor had placed them (Bolinger 1965:561). More specifically, distinguishers could be used to refer to knowledge of the world. This suggestion, abandoned later in the paper, was to be revived on several occasions (cf. e.g. Stachowiak 1982:15, and Klinkenberg 1983:298 = 1984:1173). According to Bolinger, it had the advantage of reducing two dualisms to one, but the disadvantage of leaving the remaining one intact. The author’s reasons to also reject the second dualism were as follows (Bolinger 1965:568):

The speaker’s ‘knowledge of the world’ comes in at the point where two possible theories of disambiguation are pitted against each other, one a theory (the common one) that ambiguities are resolved by the context of situation, the other that they are resolved by rules operating on markers that are part of the linguistic apparatus of a sentence. K-F do not deny the role of the nonlinguistic context in resolving many ambiguities; what they deny is that any tightly constructed theory can be built that will reflect its operation, because it involves too much; nothing less, in fact, than everything we know. The example offered to illustrate the point is Our store sells alligator shoes vs. Our store sells horse shoes. Our knowledge of the world tells us that alligators do not wear shoes, hence the first of these cannot mean ‘shoes for alligators’, but that horses do, whence the second probably does not mean shoes made of horsehide. (…) In other words, we achieve a disambiguation by way of something that is not a semantic marker.

But why is it not a semantic marker? Where do markers like (Animal), (Physical Object), (Young), and (Female) come from if not from our knowledge of the world? What is strange about (Shoe-wearing) as a semantic marker – not as general, surely, as (Female), but general enough? The discalced branch of Carmelite monks is identified by it, and it crops up every now and then as a mark of status, like horse-riding: ‘A Methodist, it was said, is a Baptist who wears shoes…’[30]

Although they do receive a mention in his writings, Katz does not appear to have taken Bolinger’s remarks as seriously as those of language philosopher Neil L. Wilson, who is virtually unknown in linguistic quarters. In his review of Katz (1966), Wilson (1967:63) observed that “there is no sharp line between what properly belongs in a dictionary and what properly belongs in an encyclopedia”. In a footnote, he acknowledged that, for this “neat formulation”, he was indebted to some “oral remarks of Katz’s”. The statement was as much a condemnation of Katz & Fodor’s conception of the dictionary as it was a gloss of Quine’s (1953) dismissal of a sharp boundary between analytic truths (which are “grounded in meanings independently of matters of fact”, and therefore – according to Wilson – linguistic in nature) and synthetic truths (which are “grounded in fact”, and therefore – again according to Wilson – encyclopedic in nature). The link between the dictionary-encyclopedia and the analytic-synthetic debates has been with us ever since (cf. Haiman 1980, Silingardi 1983:305 and Marconi 1997, among others, as well as the brief remarks in section 3).[31]

Katz (1972:73) construed Wilson’s criticism – which, as I have mentioned, he seems to have found more worthy of a reply than Bolinger’s – as an attempt to replace his dictionary entries with encyclopedia-like entries which present the “common core of factual beliefs about the referents of a word”. This he found unacceptable. He defended at length his views of the dictionary and of what dictionary entries were supposed to look like (Katz 1972:68-82), and he continued to argue that there was no need for encyclopedic information or an encyclopedic component in a semantic theory. The following passage summarises his argumentation (for full details, see Katz 1972:ch.6):

The proper question to ask is by what criterion are we to decide when a piece of information is correctly represented in the dictionary entry for a word as part of the dictionary’s account of a sense of the word and when a piece of information is correctly excluded from lexical readings. (…) The criterion can be thought of as an answer to the following question: how do we make a justified choice between two lexical readings R1 and R2 for the word W if they are exactly the same except that R2 but not R1 contains a symbol or symbols that (ex hypothesi) represent information of the sort that properly belongs in an encyclopedia entry for the thing(s) to which W refers, i.e., information of a factual and purely contingent nature about everything to which W refers? (…) The answer, roughly, is that we are to choose R2 over R1 if the incorporation of R2 in the dictionary of the semantic component as the lexical reading of W enables us to predict a range of semantic properties and relations of sentences (for example, their semantic ambiguity, synonymy, semantic anomaly, redundancy) that cannot be predicted by incorporating R1 in the dictionary in place of R2, and we are to choose R1 over R2 if the incorporation of R2 in the dictionary entry for W does not enable us to predict anything that is not already predicted on the basis of R1. What this criterion says, then, is that if the information represented in the symbol(s) that constitutes the only formal difference between R2 and R1 plays a role in predicting semantic properties and relations of sentences (…), then this information is dictionary information, information about meaning; but if we can simplify the dictionary entry for W by not including the symbol(s) that distinguish R2 from R1 without the semantic component’s losing any predictive power, then this information is not dictionary information but encyclopedia information, factual information about the referent of W. (Katz 1972:76)

It appears as though Katz has never renounced the view that there is no place for encyclopedic knowledge in a semantic description, and that the distinction between “dictionary information” and “encyclopedia information” is an essential one. Chomsky, on the other hand, did not stick to his guns – and this may come as news to many. After the ringing endorsement in Chomsky (1965) came a change of mind that has apparently never been noted, which is surprising since remarks made by Chomsky usually do not go unnoticed. Unless they are uninformative, they are quoted and interpreted by friend and foe alike, to the point where they virtually start a life of their own (cf. Peeters forthcoming b). The statement I have in mind seems to have slipped through the net. Although it is far from being uninformative, I have seen it reported only once, namely by Silingardi (1983:305), who does not refer to either Katz & Fodor (1963) or Chomsky (1965). She does refer to one of four (!) identical publications in which Chomsky recognises that “a full dictionary cannot be distinguished in a principled way from an encyclopedia” (Chomsky 1977:36; cf. also Chomsky 1975a, 1975b, 1975c). The adjective full is of course important: it indicates that, according to Chomsky, writing in the seventies, a dictionary such as Katz & Fodor’s was in fact an abridged encyclopedia, or part of an encyclopedia. Jackendoff, whose views are often close to Chomsky’s, would later suggest that information in the lexical entry “shades toward ‘encyclopedia’ rather than ‘dictionary’ information, with no sharp line drawn between the two types” (Jackendoff 1983:139-140; cf. Wilson’s formulation cited above).

The boundary problem raised by Bolinger (1965) and by Wilson (1967) has pushed some to adopt the view that, where it exists, encyclopedic knowledge is simply “more detailed” than lexical knowledge. The originality of this position, most clearly expressed in Leiss (1986:80), resides in the added comment. Leiss referred to the oft-discussed question “ob semantische Merkmale nur sprachlicher Art sind, oder ob auch sogenanntes Weltwissen, also enzyklopädisches Wissen Bestandteil der semantischen Merkmalsbündel ist”, i.e. ‘whether semantic features are exclusively linguistic in nature, or whether so-called knowledge of the world, in other words encyclopedic knowledge, may also be part of semantic feature bundles’. She then went on to say:[32]

Sprengel (1980), for instance, asks how sentences such as The elephant had eighty legs could be rejected as nonsensical. To reject such sentences, should he allow for an additional feature /+ four-legged/ in the word elephant? And how many more additional semantic features ought to be identified in the end to guarantee semantically correct utterances? The fact of the matter is that nouns, and in particular essentially referential terms, could not be adequately described without “knowledge of the world”. As a result, the entries for elephant in a lexicon and in an encyclopedia are comparable, the only difference being that the encyclopedia provides even more information. The situation is different with respect to the closed class of relational terms [on this concept, see Born-Rauchenecker, this volume; B.P.]; as a general rule, they are not to be found among the entries in an encyclopedia, even though they will be found in a lexicon (…).

Sprengel’s (1980) example of an elephant with eighty legs is strangely similar to Leech’s (1981:84) example of a dog with eighty legs.[33] Leech pleaded against the inclusion of encyclopedic knowledge in semantic descriptions: he was of the opinion that “the oddity of propositions like ‘The dog had eighty legs’ is something that zoology has to explain rather than conceptual semantics”. In Keith Allan’s work, there are at least two different comments on this statement; they show, in a subtle way, that Allan’s thoughts on the lexicon-encyclopedia interface have evolved over the years. Allan currently defends a view similar to that of Chomsky in the nineteen seventies and Leiss in the nineteen eighties (cf. below). Here is his first assessment of Leech’s claim (Allan 1992:357):

Leech is mistaken, or rather he puts the wrong case: whereas we should look to zoology to explain why genera of higher animals have no more than four limbs, we should look to linguistic semantics to recognize that, if the speaker or writer is speaking of a world identical with or similar to ours, the statement The dog had eighty legs is either false or identifying an incredibly abnormal creature. If we are interested in properly accounting for discourse coherence (i.e., in the grammar of texts larger than a sentence), then this is surely a matter that an adequate linguistic semantics should recognize – and for exactly the same kinds of reasons that it recognizes the animacy of a dog (…).

Now, here is the same author, in a paper published a few years later (Allan 1995:294):

Leech is surely correct, but there has to be path from the listeme [= lexicon entry, B.P.] to directly access encyclopedic information about dogs (or whatever) because of Hearer’s ability to ‘shadow’ a text very rapidly – that is, to begin understanding it and making appropriate inferences milliseconds after Speaker has presented it (cf. Marslen-Wilson 1985, 1989). Consequently, we might presume that

The lexicon entry is one access point into the isomorphic set of ency

阅读：268