Introduction xi

Introduction

therefore also 'cognitive' in the original sense of the word (i.e.'related to knowledge'). However, it is not the kind of 'cognitivelinguistics' that this book is about. To distinguish this approachfrom what we have in rnind when we talk about 'cognitive lin-guistics', we will characterize it as the logical view in cognitivelinguistics.

&, we understand the term, cognitive linguistics is today repre-sented by three main approaches: the experiential view, the •prorninence view and the attentional view of language. To beginwith the experiential view, its main claim is that instead of pos-tulating logical rules and objective definitions on the basis oftheoretical considerations and introspection, a more practical andempirical path should be pursued. For example, one can ask lan-guage users to describe what is going on in their rninds when theyproduce and understand words and sentences. As experiments haveshown, people will not only state that a car has a box-like shape,that it has wheels, doors, and windows, that it is driven by an engineand equipped with a steering wheel, an accelerator and brakes, andthat it has seats for the driver and the passengers. More likely thannot, it will also be mentioned that a car is cornfortable and fast, thatit offers mobility, independence and perhaps social status. Somepeople may connect the notion of car with their first love affair, orwith injury if they were once involved in an accident.

By adding these attributes, people include associations andimpressions which are part of their experience. While the last twoiterns ('frrst love affair', 'injury') point to a very personal; subjec-tive experience, attributes like 'cornfort', 'speed', 'mobility' and'independence' seem to be part of our cornmunal experience ofcars. These attributes obviously go beyond the objective descrip-tion produced by linguists schooled on logical principies. Unlikethe introspection of these linguists, the attributes collected fromlaypersons seem to reflect the way we perceive the world aroundus and interact with it. The wide and varied experience that wehave of cars is also helpful when it comes to identifying car-likeobjects that we encounter for the first time. For example, we donot hesitate to use the word car for vehicles with only three wheelsor strange-looking safari jeeps, because we can compare them withthe idea of a typical car which we have stored in our rninds.Altogether, this experiential view of words seerns to be superior tothe logical view because it provides a much richer and more naturaldescription of their meanings.

Non-objective experiential aspects of meaning do not only

Cognitive linguistics, as presented in this book, is an approach to lan-guage that is based on our experience of the world and the way weperceive and conceptualize it. However, the term has also been usedto rifer to a 'logical' view cif language, which must be distinguished

from our use of the term here.

lf someone says to you 'Our car has broken down', your reactionmay simply be to feel sorry. For the linguist, though, even such asimple utterance calls for quite an elaborate explanation. As far asthe meaning and the grarnmar of the sentence are concerned, atraditional description would try to paraphrase the meanings of thewords used; it would analyse the clause pattern (here a simple com-bination of subject and verb or predicate), and would probably goon to discuss the use of the present perfect tense.

Modern linguists have on the whole not been satisfied withthe traditional explanation of grarnrnatical structures and wordmeanings. lndeed much work in modern linguistics has beendevoted to constructing rules that would produce our initial sen-tence, but would exclude sentences like *We car has broken down or*Our car has jumped up as not well-formed. The ultimate goal ofthis endeavour has been to establish a body of logical rules forgenerating the sentences of a language that are grarnmatically cor-rect and semantically acceptable. For the individual words a set of'objective' semantic features has been proposed which would guar-antee that the words are inserted appropriately into the grarnmaticalstructure. Thus the meaning of car would be described with thesemantic features 'inanimate', 'concrete', 'movable' and 'self-propelled' .

Assuming that the logical rules and objective semantic featureswhich generate language are stored in our memory, it seerns onlynatural to claim that they are of a mental nature. This approach is

xii Introduction

emerge in experiments and personal interviews. Cognitive linguistsbelieve that our shared experience of the world is also stored inour everyday language and can thus be gleaned from the way weexpress our ideas. In order to open this mine, however, we haveto go beyond the 'logic' of clause patterns (on which the logicalview has focused) and examine figurative language, especiallymetaphors. Looking again at our initial example Our car has brokendown, it is evident that a car does not really break down just like achair collapses so that its parts come aparto Nevertheless the con-ceptual background of this expression is clear enough. Since mostof us do not know an awfullot about cars and how they work, weuse our knowledge of chairs or other equally famdiar objects col-lapsing to understand what happens when the car's engine suddenlystops working.

This transfer of our experience of well-known objects andevents is even more important where abstract categories like emo-tions are involved. Imagine that someone describes the car owner'sreaction to the breakdown of his car with the words Dad exploded.In order to get a full grasp of this utterance and the notion of angerexpressed, we will call up our knowledge of actual explosions ofgas stoves, fireworks and even bombs. This means that we willmake use of our experience of the concrete world around usoConsidering the wealth of observations, impressions and associa-tions underlying metaphors, it is not surprising that they havejoined tests and interviews as the second major basis of the experi-ential approach.

Another aspect of linguistic utterances that goes beyondlogical reasoning and objectivity concerns the selection andarrangement of the information that is expressed. For example con-sider the sentence The car erashed into the tree which might be adescription of the circumstances that led to the car's breakdown.Visualizing the accident situation sketched in this example, you willprobably agree that the sentence seems to describe the situation ina fairly natural way. In comparison, other ways of relating the acci-dent such as The tree was hit by the ear seem somehow strange andunnatural. The reason is that the moving car is the most interest-ing and prominent aspect of the whole situation, and therefore wetend to begin the sentence with the noun phrase the ear. What thisexplanation claims is that the selection of the clause subject is deter-mined by the different degrees of prominence carried by theelements involved in a situation. This prominence is not justreflected in the selection of the subject as opposed to the object

Introduction xiii.

and the adverbials of a clause, but there are also many other appli-cations of what may be called the prominence view of linguisticstructures.

The prominence view provides one explanation of how theinformation in a clause is selected and arranged. An alternativeapproach is based on the assumption that what we actually expressreflects which parts of an event attract our attention, and it cantherefore be called the attentional view. Returning once more tothe road accident, the sentence The car erashed into the tree selectsonly a small section of the event that we probably conjure up inour minds: how the CM started to swerve, how it skidded across theroad and rumbled onto the verge. Although all this happenedbefore the car hit the tree, it is not mentioned because our atten-tion is focused on the crucial point where the path of the car ended.Analysing the sentence in terms of attention allocation, the atten-tional view explains why one stage of the event is expressed in thesentence and why other stages are noto Taken together, prominenceand attention allocation seem to be no less relevant for syntacticanalysis than the rule-based description of logical grarnmars.

This book is divided up into six chapters. The first chapter willpursue the experiential view by looking at early psychological stud-ies of cognitive categories (most of them conducted by EleanorRosch), which led to the prototype model of categorization. Thiswill take us to a discussion of attributes, fami1yresemblances and ges-taltsJContrary to what one might assume, prototypes and cognitivecategories are not static, but shift with the context in which a wordis used and depend on the cognitive models stored in our mind.

The second chapter concentrates on the predominance of 'themiddle' level of categorization, called basic level. It is argued thatbasic level categories for objects and organisms, such as DOG,RABBIT or KNIFE, are cognitively more important than either super-ordinate categories like ANIMAL or CUTLERY or subordinatecategories like GREYHOUND or PENKNIFE. As it will emerge, thenotion ofbasic level categories can be transferred to the domain ofactions. or the description of properties it competes with anothercognitive notion, the image schema.

Still within the framework of the experiential view, the thirdchapter starts out from the conceptual potential of metaphors(which has already been illustrated for the breakdown of the car).This potential, which was first pointed out by Lakoff and Johnson(1980), makes a significant contribution to the cognitive contentand structure of abstract categories, especially emotion categories.

. I

xiv Introduction

This view implies that metaphors are no longer regarded as orna-mental figures of speech (as in traditional stylistics), but areunderstood as important conceptual tools. The category-structur-ing power of metaphors is not restricted to lexical categories, butcan also contribute to our understanding of complex scientific,political and social issues.

The fourth chapter is devoted to the prominence view. At theheart of this approach lies the principIe of figure/ ground segregation,which has its origin in the work on visual perception by gestalt psy-chologists. This principIe is first applied to locative relations underIyingprepositions like out or over. Then it is extended to describe other syn-tactic relations, in particular the prominence of subject versus object.The chapter ends with a rough sketch of Langacker's (1987a/1991)view of cognitive processes, which is shown to be based on a mul-tiple application of the figure/ ground contrast.

In the fifth chapter the potential of the attentional view willbe demonstrated. The chapter (which owes much to the ideas ofFillmore, Talmy and Slobin) starts out from the notion of 'frame'.Basically, a frame is an assemblage of the knowledge we have abouta certain situation, e.g. buying and selling. Depending on wherewe direct our attention, we can select and highlight differentaspects of the frame, thus arriving at different linguistic expressions ..Although elementary types of frames, for instance the 'rnotionevent-frame', are presumably shared by all human beings, they areexpressed in different ways in different languages; this will be illus-trated with English, German, French and Spanish examples.

The sixth chapter brings together a number of issues that havenot originated in cognitive linguistic research. Although three ofthem, iconicity, grarnmaticalization and lexical change, can look backon a long tradition in linguistics, they have benefited considerablyfrom being put on a cognitive basis. The [mal section discusses thepotential of a cognitive approach to foreign language learning.

To return to the general question of how 'cognitive linguis-tics' can be understood, the book will focus on the experientialaspects and the principies of prominence and attention allocationunderlying language rather than the logical view. However, as weconceive it, the experiential view provides for all kinds of experi-ences to be considered, including logical relationships. Thus thelogical element will not be denied its role in language processing,but it will be stripped of the exceptional status it holds in the'Iogical view' and will be regarded as one type of mental experi-ence beside others.

CHAPTER 1

Prototypes and categories

1.1 Colours, squares, birds and cups: early empiricalresearch into lexical categories

The world consists cif an infinite variety cif objects with different sub-stances, shapes and colours. How do we translate this variety intomanageable word meanings and why do we succeed even where noclear-cut distinctions seem to be available, such as between thecolours 'red' and 'orange' or 'green' and 'blue'? Experimentalpsychology has shown that we use Jocal or prototypical colours aspoints oJ orientation, and comparable obseruations have also beenmade with categoriesdenoting shapes, animals, plants and man-madeobjects.

Moving through the world we find ourselves surrounded by avariety of different phenomena. The most eye-catching 'arnongthem are organisms and objects: people, animals, plants and allkinds of everyday artefacts such as books, chairs, cars and houses.In normal circumstances we have no difficulty in identifying andclassifying any of them, and in attributing appropriate class namesto them. Not so easy to identify, classify and, as a consequence, toname are other types of entities, for instance parts of organisms.Knees, ankles and feet of human beings and animals or the trunk,branches and twigs of a tree belong to this type. It may be fairlyclear that one's kneecap belongs to one's knee and that the trunkof a tree includes the section which grows out of the ground. Yetat which point does one's knee end and where do es one's thighstart? Where does a trunk turn into a treetop and where does abranch turn into a twig? Similar problems arise with landscapenames and words denoting weather phenomena. Who can tell atwhich particular spot a valley is no longer a valley but a slope or a

2 An Introduction to Cognitive Linguistics

mountain? Who can reliably identify the point where drizzle turnsinto rain, rain into snow, where mist or fog begin or end?

When we compare the two types of entities mentioned, wefind that they differ with respect to their boundaries. Books, tables,cars and houses are clearly delimited objects. In contrast, theboundaries of entities like knee, trunk, valley and mist are far fromclear; they are vague. This vagueness has troubled philosophers andlinguists interested in the relationship between word meanings andextra-linguistic reality, and has given rise to various theories ofvagueness.1* Yet in spite of their vagueness we have the impres-sion that these boundaries exist in reality. A kneecap cannot beincluded in the thigh, and a mountain top will never be part of avalley. So classification seerns to be forced upon us by the bound-aries provided by reality.

However, there are phenomena in the world where this is notthe case. Take physical properties such as length, width, height,temperature and colours, all of them uninterrupted scales extend-ing between two extremes. How do we know where to draw theline between cold, warm and hot water? And how do we manageto distribute the major colour terms available in English across the7,500,000 colour shades which we are apparently able to discrim-inate (see Brown and Lenneberg 1954: 457)? The temperature scaleand the colour continuum do not provide natural divisions whichcould be compared with the boundaries of books, cars, and evenknees or valleys.

Therefore the classification of temperature and colours canonly be conceived as a mental process, and it is hardly surprisingthat physical properties, and colours especially, have served as thestarting point for the psychological and conceptual view of wordmeanings which is at the heart of cognitive linguistics. This men-tal process of classification (whose complex nature will becomeclearer as we go on) is now commonly called categorization, andits product are the cognitive categories, e.g. the colour cate-gories RED, YELLOW, GREEN and BLUE, etc. (another widely usedterm is 'concept').

What are the principles guiding the mental process of cate-gorization and, more specifically, of colour categorization? Oneexplanation is that colour categories are totally arbitrary. For a longtime this was what most researchers in the field believed. In the1950s and 1960s, anthropologists investigated cross-linguistic

Prototypes and categories 3

differences in colour naming and found that colour terrns differedenormously between languages (Brown and Lenneberg 1954;Lenneberg 1967). This was interpreted as a proof of the arbitrarynature of col~u.r ca~egories. More generally, it was thought to sup-port the relativist view oflanguages, which, in its strongest versionas a?v~cated by Whorf, assumes that different languages carve upreality In totally different ways.2

A second explanation might be that the colour continuum isstructured by a system of reference points for orientation. Andindeed, the anthropologists Brent Berlin and Paul Kay (1969)found evidence that we rely on so-called focal colours for colourcategorization. Berlín and Kay's main target was to refute therelativist hypothesis by establishing a hierarchy of focal colourswhich could be regarded as universal. To support the universalistclaim they investigated 98 languages, 20 in oral tests and the restbased on grarnmars and other written materials. In retrospect, theirtypological findings, which in fact have not remained uncriticizedhave lost some of their glamour. However, the notion of fecalcolours, which emerged from the experiments, now appears as oneof the most important steps on the way to the prototype model ofcategorization. We will therefore confine our account of Berlinand Kay's work to aspects relevant for the prototype model, at theexpense of typological details.f

Focal colours

* Suggestions for further reading are given at the end of each chapter.

Like other researchers before them, Berlin and Kay worked withso-called Munsell colour chips provided by a company of the samename. These chips are standardized for the three dimensions whichar~ relevant for our perception of different colours, namely hue,bnghtness and saturation, of which mainly the first two weretested. The advantage of using such standardized colour samplesrather than pieces of dyed cloth is that anthropological andpsychological tests become more objective, since they can berepeated by other researchers and the findings of different tests canbe compared. The set of chips used by Berlin and Kay was com-posed of 329 colour chips, 320 of which represented 40 differentcolours, or, more precisely, 40 hues, each divided up into eight dif-ferent levels of brightness. The remaining nine chips were white,?lack, and seven levels of grey. The chips were set out on a cardIII the manner shown in Figure 1.1.

4 An Introduction to Cognitive Linguistics

~ <U-o o e ::> <Ue :::l <U :D e-<U

~I >- I<U ::>~ ~ I <Ue e I e- e- e,

o o <U <U <U <U I-o :::l :::l <U <U ::> ::> ::> ::> -o<U <U <U 5h 5h :D :D o.. o.. e•... >- >-

5 10 15 20 25 30 35 40brightwhite

rgrey

1

bl-+-l++l-++++I-++++-1I-+++-H++t+t+t-t+trrt-ttttiT1

f ~rl4++~++~HH++rH~++rH~++hH;TTHg 1-+-I++l-+++HI-++++-1+++-H++t+t+t-t-ttttit-ttttiT1h LU~LU~~LU~LU~LU~LU~~~~darkblack

Figure 1.1 Arrangement of Munsell colour chips used by Berlin andKay (numbers and letters added)

The vertical axis in the figure displays the various shades ofbrightness of one identical hue. On the horizontal axis the chipsare ordered in such a way that starting from red the hues movethrough red-yellow to yellow through yellow-green to green andso on.

With the help of the colour card Berlin and Kay set abouttesting how speakers of the 20 selected languages categorizedcolours. In doing so, they were not so much interested in thecolour vocabulary in general, but rather in a particular set of colourterms which met the following criteria: the terms should consist ofjust one word of native origin (as opposed to greenish-blue andturquoise); their application should not be restricted to a narrow classof objects (as opposed, e.g., to English and German blond); thewords should come to mind readily and should be familiar to allor at least to most speakers of a language (as opposed to, say,vermilion, magenta or índigo). Colour terms which fulfilled thesecriteria were called basic colour terrns,

In the first stage of the experiments, Berlin and Kay col-lected the basic colour terms of the 20 languages. This wasachieved by means of a 'verbal elicitation test', which is just a morecomplicated way of saying that speakers of the respective lan-guages were asked to name them. In the second stage, thesespeakers were shown the colour card and asked t<?point out

Prototypes and categories 5

(1) all those chips which [they] would under any conditions call x.(2) the best, most typical examples of x.

(Berlin and Kay 1969: 7)

The questions show that, unlike Lenneberg and other anthro-pologists before them, Berlin and Kay were not only interested inthe extension of colour categories, but also in their best examples.One might even say that what was later called 'prototype' is anti-cipated in the wording of their second question.

What were Berlin and Kay's findings? In categorizing colourspeople rely on certain points in the colour space for orientation.For example, when speakers of English were asked for the bestexample of the colour 'red', they consistently pointed to colourchips in the lower, i.e. darker, regions under the label 'red' (f3 andg3 in Figure 1.1; of course, in the tests no colour terms were givenon the card). For yellow, informants consistently selected chipswith the second degree of brightness under the label 'yellow' (b9in Figure 1.1). These chips (or regions in the colour space), whichwere thought of as best examples by all or by most speakers ofEnglish, were called 'foci' by Berlin and Kay.

Foci or focal colours were also found for the other 19 lan-guages. When the focal colours were compared, the result wasamazing. Focal colours are not only shared by the speakers of oneand the same language but they are also very consistent across dif-ferent languages. Whenever a language has colour terms roughlycorresponding to the English colour terms, their focal points willbe in the same area. And even in languages with a smaller numberofbasic colour terrns than English, the best examples of these fewercategories will agree with the respective focal colours of 'richer'languages like English.

In sum, there is compelling evidence that instead of beingarbitrary, colour categorization is ancho red in focal colours. Whilethe boundaries of colour categories vary between languages andeven between speakers of one language, focal colours are shared bydifferent speakers and even different language communities.

As is often the case with important scientific findings, the dis-Covery of focal colours not only helped to solve one problem butalso raised a number of new questions: Are focal colours to betreated as a phenomenon which is a matter of language or of thernind? What, assuming the latter, is their psychological status? Andfinally: Are 'foci' (focal points) restricted to colours or can they befound in other areas as well? These questions will be taken up inthe following sections.

6 An Introduction to Cognitive Linguistics

The psychological background of focal colours

From a psychological standpoint the categorization of naturalphenomena is a rather complex task involving the following pro-

cesses:"

(1) Selection of stimuli Of the wealth of stimuli which are per-ceived by our sensory systems (visual, auditive, tactile,olfactory) only very few are selected for cognitive processing,i.e. they attract our attention.

(2) Identification and classification This is achieved by comparingselected stimuli to relevant knowledge stored in memory.

(3) Naming Most cognitive categories are given names thoughsome remain unlabelled, e.g. 'things to eat on a diet', 'thingsto pack in a suitcase' (Barsalou 1987: 102).

Most of these aspects were investigated by Eleanor Rosch,who in the early 1970s set out to explore the psychological back-ground of focal colours.P As a psychologist, her primary aim wasto find out whether focal colours were rooted in language or inpre-linguistic cognitiorf Her idea was that a cognitive status rnightbe claimed for focal colours if they could be proved to be prornin-ent in the cognitive p¡;-ocessesinvolved in categorization.

Starting out from the most basic of the three cognitive pro-cesses, Rosch first exarnined whether focal colours are perceptuallysalient. To elirninate the influence of purely language-based cate-gorization, she required informants who had stored as littleknowledge of colour names and related colour categories as pos-sible. So she decided to work with pre-school children and withmembers of a non- Westernized culture in Papua New Guinea, theDani. Earlier research had shown that Dugum Dani, the languagespoken by the Dani, contained only two basic colour terms, incontrast to the 11 basic colour terms available to speakers ofEnglish(Heider 1970). Like children, the Dani were therefore particularlywell suited as uncorrupted informants for colour categorizingexperiments. English-speaking adults, who were supposed to havethe full system of basic colour terms at their disposal, were onlyused as control groups in some of the tests.

Rosch's first experiment (Heider 1971), which was to test thearousal of attention (or stimulus selection), was dressed up as a'show me a colour' game. She gave 3-year-old children arrays ofcolour chips consisting of one focal colour, as found by Berlin andKay, and seven other chips of the same hue, but other levels of

r

•

Prototypes and categories 7

brightness. The children were told that they were to show theexperimenter any colour they liked. The reasoning behind thisgarne was that young children's attention would be attracted morereadily by focal colours than by other colours. In fact, it turned outthat the children did pick out focal chips more frequently thannon-focal chips. The preponderance of the focal chips was parti-cularly strong for the colours yellow, orange and green, where 22,21 and 11 respectively out of the total of 24 children selected thefocal chip from the array. For the other five hues, the numberswere smaller, but stiU statistically significant.

The second experiment which Rosch conducted with chil-dren was a colour-rnatching task. The children, this time4-year-olds, were given focal and non-focal chips one at a time inrandom order and asked to point to the same colour in an array ofcolour chips which were identical to those used in the earlierexperimento As predicted by the test hypothesis, focal colour chipswere matched more accurately than non-focal chips, and this agains~~ports the perceptual salience of focal colours. In terms of cog-mtrve processes, this second test involves identification andcla.ssification; both the test chip and one or several possible targetchips have to be identified and classified so that they can be com-pared. Comparison in turn presupposes that the data collectedabout the chips are temporarily stored somewhere, and this iswhere memory comes into play.. Matc~ng situations where both the test item and the targetitems are simultaneously present are rather the exception. Normallywe are confronted with an item (i.e. a colour which has to be iden-tified and classified) but have to rely fully on data stored in memoryfor comparison! This raises the question whether focal colours aresalient in memory as well, whether they are recognized moreaccurately, learned more easily and recalled more readily than othercolours. Investigating these aspects Rosch used specific memorytests: r~cognition tasks to test the short-term memory and learningtasks aimed at the long-term memory (Heider 1972).. . The recognition task (addressing the short-term memory) was

sunilar to the matching task discussed above, but demanded moreco?centration from the informants. As this proved too difficult forchildren, Rosch used Dani informants who, as already mentioned,have only two basic colour terms. The same test was conductedwith ~ co~trol group of Americans. Both the Dani and the English-speaking informants were shown eight focal and eight non-focalcolour cards in random order each for five seconds. Each single

••

8 An Introduction to Cognitive Linguistics

presentation was followed by an interval of 30 s~conds, after ~hichthe subjects were asked to point out the matching colour chip ona colour card of the Berlin and Kay type. For both groups thematching accuracy for focal colours was significantly higher thanfor non-focal ones. The conclusion one may draw is that focalcolours are remembered more accurately in short-term memorythan non-focal ones. Another result was that the English-speakingcontrol group surpassed the Dani in matching accuracy both forfocal and non-focal colours. This may be due to the fact that thelarger set ofbasic colour terms available to speakers ofEnglish facil-itates colour recognition.

The second experiment consisted of a learning task testingretention of previously unknown colour names in long-termmemory. This experiment exploited the fact that, because of theirlimitation to two basic colour terms, the Dani could be taughtadditional ones under controlled conditions. Before the test began,the Dani informants were told that the experimenter would teachthem a new language. When learning was completed, they werepaid for their help. At the start, the 16 colour cards (again eightfocal colours and eight non-focal colours) were laid out in randomorder and the Dani were told the name for each cardo (The namesused were the names of Dani clans.) Mter their first display thecards were gathered into a pack, shufRed and presented one by oneto the Dani, who were asked to produce the name of each colour.They were praised for every correct response and told the correctname when they were wrong. This procedure was repeated fivetimes a day until the Dani managed to get all 16 answers right andthe learning process was considered completed. A detailed recordwas kept of the whole learning process, which took three and ahalf days on average. This record supplied the means of measuringthe ease of retention of focal and non-focal colours beca use itallowed a computation of all the correct and incorrect answers.Whereas, on average, the Dani gave 9.9 incorrect answers percolour for non-focal colours before they produced their first com-pletely correct run, the mean number of errors per colour for focalcolours was only 7.3. Even without previous knowledge of thecolour names, the Dani associated focal colours more rapidly withtheir names than non-focal colours.

As well as perception and memory work, narning was men-tioned as the third component of categorization. Given the salienceof focal colours we would expect the following results: first, namesshould be produced more rapidly for focal colours than for

Prototypes and categories 9

non-focal ones; second, children should acquire the names of focalcolours earlier than the names of non-focal colours. Empirical evid-ence, again provided by Rosch, suggests that both assumptions arecorrecto

Let us now review Rosch's findings:

(1) Focal colours are perceptually more salient than non-focalcolours. The attention of 3 year olds is more often attractedby focal than by non-focal colours, and 4 year olds matchfocal colours more accurately to a given display of othercolours than non-focal colours.

(2) Focal colours are more accurately remembered in short-termmemory and more easily retained in long-term memory.

(3) The names of focal colours are more rapidly produced incolour-narning tasks and are acquired earlier by children.

All in all, focal colours appear to possess a particular percep-tual-cognitive salience, which is probably independent oflanguage.In view of research by Kay and McDaniel (1978) we rnight evensay that the salience of focal colours reflects certain physiologicalaspects of man's perceptive mechanisms. Another line of explana-tion relates focal colours to certain universally occurringphenomena like day and night (white and black), the sun (red),vegetation (green), the sky (blue) and the ground (brown)(Wierzbicka 1990).

However, impressive as the findings on the salience of focalcolours may be, they do not in themselves solve the problems ofcolour categorization. What is equally important is to define thefunction of focal colours in the formation of colour categories. Arefocal colours only salient if they are in the centre of a series ofsimilar colour chips? Or are they also salient if they have a mar-ginal position in a given set? If this were the case, it would seemthat focal colours are used as 'anchors' for their colour categoriesirrespective of their position in the set, i.e. that focal 'red' wouldalways be used as an 'anchor' for all kinds of redness.

In order to test this hypothesis Rosch (1973) replaced the iso-lated colour chips that she had used in her previous experimentswith sets of three colour chips, some with the focal colour in a cen-tral position, some with the focal colour in a marginal position. Indoing so, she deliberately constructed artificial combinations ofcolour chips. The same method was also applied to geometricalshapes, and we will explain the underlying principIe when we dis-cuss shapes in the next section. With the new set-up Rosch

10 An Introduction to Cognitive Linguistics

repeated the memory tests and found that focal colours weresalient no matter where they were located within a given seto

This result had an important terminological consequence.Since the label 'focal' suggests a central position, the term was nolonger appropriate, and this is probably one of the reasons whyRosch replaced Berlin and Kay's 'focus' with prototyp!.: The termwas borrowed from earlier research into pattem recognition of suchstirnuli as dot patterns, polygons and schematic faces (see, e.g., Reed1972). In the context of this research, 'prototype' stood for artifi-cially created 'best examples'. Compared with these artificialprototypes, focal colours were, of course, much less arbitrary pheno-mena and were therefore called 'natural prototypes' by Rosch. Oneof the advantages of using the term 'prototype' was that it made itmuch easier to extend the notion of foci beyond colour categories,e.g. into the domains of shapes, organisms and objects.

Prototypes and categories 11

Set 1

(a) (b) (e) (d) (e) (f) (g)

DCODOOD(a) (b) (e) (d) (e) (f) (g)

S"'CDCCCOC(a) (b) (e) (d) (e) (f) (g)

S'''OCDOOOQFigure 1.2 Shapes used by Rosch in prototype experiments (Rosch1973) (Sets 2 and 3 rcconstructed from Rosch's description, letters added)

Prototypical shapes

Let us start our discussion of shapes with a little experiment basedon the drawings in the top row ofFigure 1.2. (set 1). Imagine youwere asked to describe what you see in this figure to someone whois not allowed to inspect it. Presumably you would proceed moreor less like this: 'There is a row oflittle drawings depicting a squareand a number of variations of it. The first drawing is a propersquare. The second square has a gap in the right-hand side. Theright-hand side of the third square has an indentation. The fourthsquare ... .'

Such a description would be in full agreement with thenotion of'good forms' as proposed by gestalt psychology (a schoolof psychology to which we will retum in section 1.2). These goodforrns, i.e. squares, cirdes and equilateral triangles, are assumed tobe perceptually salient among geometrical shapes.

Thus it is only natural that in a situation like the one describedabove people will single out the square as a reference point forcharacterizing the other drawings. Squares and the other goodforms are therefore prime candidates for 'natural' prototypes in thedomain of geometrical shapes, similar to the focal colours in colourspace.

Using the kind ofline drawings shown in Figure 1.2, Rosch(1973) sought confirmation for the notion of natural prototypes inthe domain of shapes. Prior to the actual tests, she had to makesure that, a with colours, the Dani had no category names or even

conventional paraphrases at their disposal which could bias themtowards the supposed prototypes. This was verified in a pilot study,which used the description method demonstrated above: one testsubject explaining the line drawings to another subject who wassitting behind a screen and could not see them. U nlike educatedWestem speakers, the Dani did not talk. of squares and variations,but used expressions like 'It's a pig' or 'It's a broken fence' for theirdescription of the drawings.

For the actual experiment, Rosch contrasted set 1 (the setwith the natural prototype) with other sets which were derivedfrom the variations of set 1. Sets 2 and 3 in Figure 1.2 show twoof the possible six altemative sets. Set 2 is based on the principleof gapping (gap on the right-hand side). Considering this principle,the natural prototype which appears as (b) in set 2 is indeed a veryextreme case because it represents the absence of the gap. Set 3 isbased on the principle of indenture, and again the square is just amarginal member of the seto

As in the earlier colour-learning experiments, the Dani hadto learn names which were again borrowed from Dani dans. Thismeans they had to associate sets of drawings with names. The resultfutiy confirmed Rosch's assumption that the natural prototype isassociated with a name and also judged best example, no matterwhether it is presented in a natural category (set 1) or as a marginalrealization of the principle underlying one of the other sets.

10 An Introduction to Cognitive Linguistics

repeated the memory tests and found that focal colours weresalient no matter where they were located within a given seto

This result had an important terrninological consequence.Since the label 'focal' suggests a central position, the term was nolonger appropriate, and this is probably one of the reasons whyRosch replaced Berlin and Kay's 'focus' with prototype. The termwas borrowed from earlier research into pattem recognition of suchstimuli as dot patterns, polygons and schematic faces (see, e.g., Reed1972). In the context of this research, 'prototype' stood for artifi-cial1y created 'best examples'. Compared with these artificialprototypes, focal colours were, of course, much less arbitrary pheno-mena and were therefore called 'natural prototypes' by Rosch. Oneof the advantages of using the term 'prototype' was that it made itmuch easier to extend the notion of foci beyond colour categories,e.g. into the domains of shapes, organisrns and objects.

Prototypical shapes

Let us start our discussion of shapes with a little experiment basedon the drawings in the top row ofFigure 1.2. (set 1). Imagine youwere asked to describe what you see in this figure to someone whois not allowed to inspect it. Presumably you would proceed moreor less like this: 'There is a row oflittle drawings depicting a squareand a number of variations of it. The first drawing is a propersquare. The second square has a gap in the right-hand side. Theright-hand side of the third square has an indentation. The fourthsquare .... '

Such a description would be in full agreement with thenotion of'good forms' as proposed by gestalt psychology (a schoolof psychology to which we will retum in section 1.2). These goodforrns, i.e. squares, circles and equilateral triangles, are assumed tobe perceptually salient among geometrical shapes.

Thus it is only natural that in a situation like the one describedabove people wil1 single out the square as a reference point forcharacterizing the other drawings. Squares and the other goodforms are therefore prime candidates for 'natural' prototypes in thedomain of geometrical shapes, similar to the focal colours in colourspace.

Using the kind ofline drawings shown in Figure 1.2, Rosch(1973) sought confirmation for the notion of natural prototypes inthe domain of shapes. Prior to the actual tests, she had to makesure that, as with colours, the Dani had no category names or even

Prototypes and categories 11

Set 1

(a) (b) (e) (d) (e) (f) (g)

DCODOOD(a) (b) (e) (d) (e) (f) (g)

S"'CDCDCOC(a) (b) (e) (d) (e) (f) (g)

S"'OCDODO[]Figure 1.2 Shapes used by Rosch in prototype experiments (Rosch1973) (Sets 2 and 3 reconstructed from Rosch's description, letters added)

conventional paraphrases at their disposal which could bias themtowards the supposed prototypes. This was verified in a pilot study,which used the description method demonstrated above: one testsubject explaining the line drawings to another subject who wassitting behind a screen and could not see them. Unlike educatedWestem speakers, the Dani did not talk of squares and variations,but used expressions like 'lt's a pig' or 'lt's a broken fence' for theirdescription of the drawings.

For the actual experiment, Rosch contrasted set 1 (the setwith the natural prototype) with other sets which were derivedfrom the variations of set 1. Sets 2 and 3 in Figure 1.2 show twoof the possible six altemative sets. Set 2 is based on the principleof gapping (gap on the right-hand side). Considering this principle,the natural prototype which appears as (b) in set 2 is indeed a veryextreme case because it represents the absence of the gap. Set 3 isbased on the principIe of indenture, and again the square is just amarginal member of the seto

As in the earlier colour-learning experiments, the Dani hadto leam names which were again borrowed from Dani clans. Thismeans they had to associate sets of drawings with names. The resultful1y confirrned Rosch's assumption that the natural prototype isassociated with a name and also judged best example, no matterwhether it is presented in a natural category (set 1) or as a marginalrealization of the principle underlying one of the other sets.

n

12 An Introduction to Cognitive Linguistics

Combined with the findings from the earlier colour experi-ments, these results suggest that natural prototypes have a crucialfunction in the various stages involved in the formation and learn-ing of categories.

Prototypes and categories 13

categoryrank BlRD FRUIT VEHICLE FURNITURE WEAPON

top eight1 robin orange automobile chair gun2 sparrow apple station wagon sofa pistol3 bluejay banana truck couch revolver4 bluebird peach car table machine gun5 canary pear bus easy chair rifle6 blackbird apricot taxi dresser switchblade7 dove tangerine jeep rocking chair knife8 lark plum ambulance coffee table dagger

middle ranks26* hawk tangelo subway lamp whip27 rayen papaya trailer stool ice pick28 goldfinch honeydew cart hassock slingshot29 parrot fig wheelchair drawers fists30 sandpiper mango yacht piano axe

last five51* ostrich nut ski picture foot52 titmouse gourd skateboard closet car53 emu olive wheelbarrow vase glass54 penguin pickle surfboard fan screwdriver55 bat squash elevator telephone shoes* Since the total number of listed items varied between 50 and 60,the nurn-

bers of middle and bottom ranks are not identical with the original ranks forall categories.

Figure 1.3 A selection of examples from Rosch's goodness-of-exarnplerating tests (Rosch 1975)

Prototypical organisms and objects

It could still be argued at this point that prototypes ultimatelydepend on the perceptual nature of the categories examined so far(colours, shapes) and are therefore a very limited if not exceptionalphenomenon. The question is whether the notion of prototype canbe extended to entities which are less obviously perceptual.Granted there are good and bad examples of reds and squares. Arethere also good and bad examples of dogs, cars and houses?According to Rosch and her informants there are. In a series ofexperiments (Rosch 1973, improved version 1975) she confrontedinformants, this time American college students, with the follow-ing test instructions:

This study has to do with what we have in mind when we use wordswhich refer to categories. Let's take the word red as an example. Closeyour eyes and imagine a true red. Now imagine an orangish red ...imagine a purple red. Although you might still name the orange red orthe purple red with the term red, they are not as good examples of red (asclear cases of what red refers to) as the clear 'rrue' red. In short, some redsare redder than others. The same is true for other kinds of categories.Think of dogs. You all have some notion ofwhat a 'real dog', a 'doggydog' is. To me a retriever or a German shepherd is a very doggy dogwhile a Pekinese is a less doggy dogo Notice that this kind of judgmenthas nothing to do with how well you like the thing. [ ... 1

(Rosch 1975: 198)

In the remainder of the instructions the students were asked tojudge the goodness (or typicality) of category members, i.e. todecide how good an example of the category BIRD a sparrow, aparrot, a penguin and about 50 other candidates were. Rating wasbased on a 7-point-scale of goodness (one point for very good,seven points for very poor examples). Altogether ten categorieswere tested: in addition to BIRD, the categories were FRUIT,

VEHICLE, VEGETABLE, SPORT, TOOL, TOY, FURNITURE, WEAPON andCLOTHING.

Admittedly, Rosch's test instructions show a certain biastowards the notion of prototypicality (' ... some reds are redderthan others. The same is true for other kinds of categories'). Yet

this should not have distorted the test results too mucho As Roschstresses, the rating test was readily accepted by the student infor-mants and there was a high level of agreement among them as towhat were good and bad examples of the categories. To give animpression of what the results were like, the best, some intermedi-ate and the poorest examples of five out of the ten categories areassembled in Figure 1.3.

The goodness ratings were also confirmed in matchingexperiments in which the 'priming' technique was used (Rosch1975). In one of these tests, subjects were shown pairs of names orpictures on a screen. The subjects had to press a 'same' key whenpairs of identical names or pictures of items appeared on a screen(e.g. a word sequence like eagle-eagle or two identical pictures of

14 An Introduction to Cognitive Linguistics

an eagle); the time between the presentation and the reaction wasmeasured (it was in the range of 500 to 1000 milliseconds). Twoseconds before the presentation, subjects were given the categoryname (in this case birá) as advance information, so they were'primed' with the category name. (There was also a control groupthat performed the test unprimed.) The hypothesis was thatadvance knowledge of the category name would influence thespeed with which the matching task was performed and that itmight influence the matching of good and poor examples in dif-ferent ways. Indeed, priming had a twofold effect which nicelysupported the goodness ratings. Primed informants were faster inreacting to identical pairs of items that had be en rated as goodexamples (both words and pictures). Conversely, reaction wasslowed down by priming where poor examples were involved.Reaction to pairs of intermediate examples was not noticeablyaffected by priming. Without getting lost in speculations about thecognitive representations of categories at this point, we can stillsupport Rosch's claim that the advance information which is calledup by the primed category name is most readily applied to goodexamples; sparrows, oranges or cars (automobiles in AmericanEnglish) imply fit the expectations called up by the names of thecategories BlRD, FRUIT and VEHICLE. However, this advance infor-mation is not helpful with poor examples. In fact, priming withthe category name tends to confuse test subjects when they areconfronted with pairs of penguins, olives or wheelbarrows, whichcan at best be placed at the periphery of the bird, the fruit or thevehicle category.

Good examples, bad examples and categoryboundaries

As the categorization of colours, shapes, birds and vehicles suggests,category membership is not, as was for a long time assumed byphilosophers and linguists, a yes-or-no distinction. Rather itinvolves different degrees of typicality, as is supported by good-ness-of-example ratings, recognition, matching and learning tasks.

Rosch's main concern was to prove that categories are formedaround prototypes, which function as cognitive reference points.As far as the boundaries of categories are concerned, she leaves uswith the impression that at some unspecified point or area beyondtheir periphery the categories somehow fade into nowhere. Thisis not the idea we have when we talk about categories in a naive

Prototypes and categories 15

way. Normally, we tend to imagine them as boxes, drawers orsome sort of fenced compound - certainly as something which hasboundaries. With regard to the category BIRD, the allocation ofboundaries seems to be easy enough, even though a little know-ledge of zoology might be required. Yet our confidence will beundermined when we follow the philosopher Max Black and con-sider the imaginary 'chair museum' he invented. According toBlack it consists of

a series of'chairs' differing in quality by least noticeable amounts. At oneend of a long line, containing perhaps thousands of exhibits, rnight be aChippendale chair: at the other, a small nondescript lump of wood. Any'normal' observer inspecting the series finds extreme difficulty in 'draw-ing the line' between chair and non-chair.

(Black 1949: 32)

What Black's interpretation ofhis chair museum suggests is that thecollection of chairs could and should be regarded as a continuumwith a kind of transition zone between chairs and non-chairs butno clear-cut boundaries. This view seems to be in conflict withwhat we observed at the beginning of the chapter: that concreteobjects like houses, books and also chairs are clearly delimited andeasy to identify, and that vague boundaries and transition zones arerestricted to items like knees, fog and valleys and to scales likelength, temperature and colour.

Here we must be careful not to confuse two different typesof boundaries and transition zones. One type of transition zonearises from the observation that some concrete entities do not haveclear-cut boundaries in reality - this is the case with knee andother body parts, it applies to fog, snow and similar weatherphenomena and to landscape forms like valley or mountain. InBlack's chair museum, however, the visitor is confronted with adifferent type of transition zone, since each exhibit in the museumis an entity with absolutely clear boundaries. In the chair museum,it is not entities that merge into each other, but categories ofentities, and these categories are the product of cognitive classifi-cation. Consequently, it is not the boundaries of entities that arevague, but the boundaries of these cognitive categories (here:chairs and non-chairs). To distinguish the two types of vaguenesswe will restrict the terrns 'vague entity' and 'vagueness' to the firsttype (knee, fog, valley) and use 'fuzzy category boundaries' orfuzziness for the second, i.e. for the category boundaries ofCHAIR, etc.

16 An Introduction to Cognitive Linguistics

The issue is, however, even more complicated beca use thereare in fact cases where vagueness and fuzziness coincide. This istrue of the second type of entities, as already observed by the philo-sopher Willard Quine, who found that the category 'mountain' is

vague on the score of how much terrain to reekon into eaeh of the indis-putable mountains, and it is vague on the seore of what lesser ernineneesto eount as mountains at all.

(Quine 1960: 126)

In other words, entities like mountains are vague because they arenot clearly delirnited as individual entities; the cognitive categoryMOUNTAIN (or KNEE or FOG) is fuzzy because it does not have clearboundaries either.

Sumrning up, we can say that our deeper understanding ofthe cognitive background of eategorization has considerablychanged our original idea about the threefold classification ofentities into clearly delirnited organisms and objects, into entitieswith vague boundaries and into scales - a view which is based ona kind of 'naive realism'. 6 From a cognitive perspective thesedistinctions and the discussion of vagueness arising from them areof minor importance. What is important is that all types of concreteentities and natural phenomena like colours are conceptuallyorganized in terms of prototype categories, whose boundaries donot seem to be clear-cut, but fuzzy.

How can the fuzzy nature of category boundaries, whichintuitively seems to be a convincing notion, be investigatedempirically? This was the task which William Labov set himself ina series of experiments involving cups and cup-like containers(Labov 1973, 1978).7 Starting from Black's interpretation of thechair museum, Labov drew the following conclusion:

The subjeetive aspeet of vagueness [i.e. fuzziness in our terrninology] maybe thought of as the laek of eertainty as to whether the term does or doesnot denote; and this may be transformed into the eonsisteney with whieha given sample of speakers does in faet apply the termo

(Labov 1973: 353)

If all informants in a test call an object chair, the consistencyis 100 per cent. If half the informants have doubts whether a cer-tain object is still a chair and therefore do not call it chair, theconsistency value will drop to 50 per cent. If hardly any of theinformants regard an object as a chair and refuse to call it a chair,the consistency value will approach zero.

Prototypes and categories 17

Figure 1.4 A seleetion of the drawings of eup-like objeets used byLabov (1973: 354); (no. 5 reeonstrueted; alternative drawings without a

handle or with two handles were also used in the tests)

. The actual test procedure of Labov's experiments was verysimple: mformants were shown line drawings of cups and othervessels, as collected in Figure 1.4. The drawings were presentedone by one and the informants were asked to name them (addi-tional descriptive details supplied by the informants were neglectedin the analysis).

The results of the naming task were analysed in terms of con-sistency and presented as 'consistency profiles'. Figure 1.5 presentsthe consistency profile for the vessels shown in the top row ofFigure 1.4.

As the graph for the use of cup indicates, consistency is 100 percent for vessel no. 1 but decreases as we proceed towards vesselno. 5. In addition, Figure 1.5 also contains the complementary graphfor the use of bowl. This graph demonstrates that Black's chairmuseum was rather unrealistic in that he only compared 'chairs' with'non-chairs'. In contrast, Labov's test shows that in experimental and

%

100cup

75

50

25

0-r-----,,------.----~~----_.~Vessel no. 2 3 4 5

Figure 1.5 Consisteney profile for neutral context (adapted from Labov1973)

18 An Introduction to Cognitive Linguistics

everyday categorizing situations, we normally do not just make adivision between the two categories 'X' and 'not X', but that wehave two or several names at our disposal which allow us to choosebetween neighbouring categories, in this case between CUP, BOWL,MUG and VASE, etc. Therefore, it is more realistic to think of fuzzycategory boundaries as fringe areas between adjacent categories thanas transitions to a conceptual vacuum. Figure 1.5 shows that as theconsistency value for CUP drops dramatically for vessels 4 and 5, theuse of bowl slowly begins to pick up. It is for these fringe areasbetween the two categories that the term 'fuzzy boundaries' seemsto be particularly appropriate. Labov's tests can therefore be takenas a first experimental proof of the fuzziness of category boundaries.

However, this fuzziness assumes a new dimension when weconsider the full range of Labov's experiments. In the first test(which we have concentrated on so far) the informants were onlyconfronted with the drawings, but not given any backgroundinformation (this was called 'neutral context' by Labov). In the sub-sequent three tests they were asked to imagine one of threedifferent scenes: (a) a coffee-drinking situation, (b) a dinner tablesituation with the object filled with mashed potatoes ('food con-text') and (c) a scene where the objects were standing on a shelfwith cut flowers in thern. In later experiments different materialslike china and glass were introduced as well.

The result of including these variables was a massive shift ofcategory boundaries. To give just one example, in a food context,vessel no. 3 was no longer a cup for the majority of the informants.As indicated in Figure 1.6, half the informants called it bowl in spite

%

100

3 4 5Vessel no. 2

Figure 1.6 Consistency profile for neutral and food contexts (Labov1973)

Prototypes and categories 19

of its unchanged shape, and this switch towards bowl was evenmore pronounced for vessel no. 4. In this way Labov's experimentsshow that the fuzziness of category boundaries has many facets, ofwhich context-dependence is one of the most important. (Thisissue will be taken up in section 1.3; another aspect ofLabov's tests,his carefully controlled use of scalar properties like width, depthand shape, will be discussed in the next chapter.)

Let us now relate Labov's findings to what has alreadyemerged about the nature of cognitive categories in the precedingsections:

(1) Categories do not represent arbitrary divisions of the pheno-mena of the world, but should be seen as based on thecognitive capacities of the human mind.

(2) Cognitive categories of colours, shapes, but also of organismsand concrete objects, are anchored in conceptually salient pro-totypes, which playa crucial part in the formation of categories.

(3) The boundaries of cognitive categories are fuzzy, i.e.neighbouring categories are not separated by rigid boundaries,but merge into each other.

(4) Between prototypes and boundaries, cognitive categories con-tain members which can be rated on a typicality scale rangingfrom good to bad examples.

If we accept that cognitive categories consist of prototype, goodexamples and bad examples, and have fuzzy boundaries, this sug-gests that the internal structure of categories is indeed rathercomplex and that it deserves a more detailed examination.

We would like to conclude with a cautioning remark.Cognitive categories, as we have discussed them in this section, arestored in our mind as mental concepts and signalled by the wordsof a language, so one might come to think that they are equiva-lent with the meanings of these words. Yet if we consider thatcolour terms do not just denote colours, but can also stand for poli-tical parties, that bird does not only refer to a creature with wingsbut, at least occasionally, also to a pretty girl, that chair can denotethe president of a meeting and that cups and bowls can be trophiesin sport, it is clear that there is no one-to-one relation betweencategories (or concepts) and words. In fact it is quite normal thatone word denotes several categories, or in conventional linguisticterrninology, that words are polysernous.f As the later chapters willshow (especially Chapters 3 and 4), cognitive linguistics is not onlyconcerned with the exploration of individual categories, but it also

/

20 An Introduction to Cognitive Linguistics

has something to say about the relationship between the cate-gories which are signalled by one and the same word.

Exercises

1. List basic colour terms in English (or your native language)by checking which colour terms consist of only one shortword and are freely applicable to different kinds of objects andorganisms. Describe the colour of sweaters, T-shirts, etc.,using basic colour terms as points of reference.

2. Select typical examples of the categories NEWSPAPER,JOURNAL, MAGAZINE and PERIODICAL. Do you know publi-cations which illustrate the fuzziness of the boundariesbetween these neighbouring categories?

3. Draw pictures of prototypical examples and of objects on theborderline between the categories BOTTLE, GLASS, VASE andBOWL, and use them as stimuli for a narning task with yourfriends or farnily.

4. As we have found, the vagueness of objects and the fuzzinessof categories must be kept aparto Look at the followingexamples and discuss which of them involve fuzziness orvagueness or both aspects:

mountain, hiil, sumrnit, plateau, valley;tree, shrub, fl.ower;hedge, bush, forest, park;street, road, avenue, drive, highway;river, stream, brook, torrent, firth, estuary, spring.

5. The names of category prototypes tend to come to rnindbefore those of peripheral examples. Check this hypothesiswith two informal tests: ask one group of friends to name asquickly as they can five types of dogs, birds, trees and cars,and a second group to rate these example for their goodness-of-example status within the category. Compare the resultsand discuss reasons for discrepancies between the two tests.

1.2 The internal structure of categories: prototypes,attributes, family resemblances and gestalt

1j cognitive categories are made up oj prototypes and periphery, of goodand bad examples, how do these differ and how are they related to

Prototypes and categories 21

each other? The listing and the analysis of attributes seem to providea good approach to these aspects oj internal category structure, whilethe notion of jamily resemblances is helpjul as a theoretical explana-tion. A jascinating though less well exp/ored jactor in categorizationis the 'gestalt' of organisms and concrete objects, which will a/so comeup jor discussion,

Cognitive categories are, as we have just seen, labeiled by words,and words are listed in dictionaries. It is therefore only natural tolook for information about the contents of categories in dictionaryentries. Here are some examples of dictionary defmitions for typesofbirds: .

robin A small brownish bird with red breast-feathers (alsocalled red-breast). (OALD)

parrot A tropical bird with curved beak and usuaily brightlycoloured feathers. Some can be taught to copy humanspeech. (LDOCE)

ostrich An ostrich is a large African bird that cannot fl.y. It haslong legs, a long neck, a small head and large softfeathers. (COBUILD)

In terms of categorization, these dictionary definitions yield twotypes of information. To start with, they supply the name of thecategory to which the robin, the parrot and the ostrich belong (inthis case bird). This category name in turn suggests the propertieswhich are shared by most birds: that they have feathers, two legs,two wings and a beak, and that they lay eggs. The main body ofthe dictionary entries lists properties which are specific to the itemin question. Thus the robin is characterized by small size, brown-ish colour and red breast-feathers. These properties clearly set therobin apart from other members of the category BIRD, such asparrots and ostriches. So robins, parrots and ostriches have proper-ti es which serve to tie them to a common category as well asproperties which distinguish them from each other. Collectingboth the shared and the distinctive properties seems to provide afeasible way of describing the internal structure of categories.

However, there are some problems. Dictionary definitions arewritten for a practical purpose and not with a systematic linguisticand cognitive analysis in rnind. Lexicographers can afford to skipsorne properties that are to be taken for granted, or they canmodify their definitions by lirniting expressions or 'hedges', like~Isually (parrots 'usually' have brightly coloured feathers). A more

22 An Introduction to Cognitive Linguistics Prototypes and categories 23

'can By''is small and lightweight',chirps/ sings'

(8) 'has thin, short legs'(9) 'has a short tail'

(10) 'has a red breast'

(5)(6)(7)

systematic linguistic approach not only has to fill in gaps, e.g. byadding to the definition of robín that it chirps and to the defmitionof ostrich that it can run very fastoA linguistic analysis will also haveto clarify the notion of property or attribute (to use the moretechnical term).

Regarding the attributes used in dictionary entries, it may bequite sufficient to understand them in the rather vague sense of'characteristics' or 'typical aspects'. From a more theoretical stance,an additional question must be asked: Are attributes to be regardedas obligatory or not? This distinction was first suggested byAristotle, who contrasted the 'essence' of things with the 'acci-dence'. The notion of essence gave rise to what we have called thelogical view (other terms used in the literature are 'categoricalview' or 'classical view'), a position which was vigorously defendedby structuralist and transformationalist linguists.i' According to thisview, a category is defined by a lirnited set of necessary and suffi-cient conditions. These conditions are conceived as olear-cut,'discrete' features (or essential features, as we will call them),which can be either present or absent. In the case of the categoryBIRD, this means that a creature is only a bird if it has two wingsand two legs, a beak, feathers and lays eggs (these are the necessaryconditions). If, on the other hand, a creature has all these essentialfeatures, this is also sufficient for classifying it as a bird.

Such a rigid view of attributes and categorization is bound torun into difficulties when it is applied to cognitive categories con-sisting of good and bad examples and equipped with fuzzyboundaries, which we introduced in section 1.1. To cope withthese experiential prototype categories we need a much more dif-ferentiated notion of attributes than is supplied by the logical view.The following discussion of attributes for birds will exemplify theproblems.

The list is based on dictionary definitions of robín and bird, but alsosupported by attributes collected from informants (Hampton 1979).More precisely, it assembles what dictionary makers and ordinarypeople (the informants) fmd worth mentioning about robins. Sucha list will never be complete (for example, our description does notmention what kind of food a robin eats) and the items will tend tooverlap (in our case this applies to nos 3 and 8, which both referto the bird's legs, and to nos 4 and 10, which are both concernedwith the bird's plumage). Yet while these deficiencies may be con-fusing to the linguist raised in the classical tradition, they still seem,for the time being, to provide the best empirical way of describ-ing the properties which we use in categorizing a robin.

If we try to apply the attributes collected for robins to otherexamples of the category BIRD which scored high in Rosch's rat-ing test, e.g. >SPARROW<, >CANARY< and >DOVE<, we find that,with the exception of the red breast, sparrows and canaries shareall the attributes we have assembled for robins. Doves do not chirpor sing and surpass the other three types ofbirds in size and weight;otherwise they share the attributes listed. It seerns, therefore, thatat least the attributes 'lays eggs', 'has a beak', 'chirps/sings', all theattributes concerning wings, feathers and the ability to By, and,finally, the 'thin, short legs' and the 'short tail' are sornehow relatedto a central position within the category BIRD.

Now take a less good example of a bird, say >PARROT<. Ifwe go through our list again, a parrot resembles a robin in that itlays eggs, has a beak, two wings, two legs and feathers and can By.However, compared to a robin a parrot has rather strong legs, mostparrots are much larger than robins and have quite long tails. Anda parrot certainly do es not chirp or sing.

Finally, when we consider the attributes of >OSTRICH< (whichwas, of course, rated a poor example of the category BIRD), we findthat the only attributes it shares with our short list are that it layseggs, has two legs and feathers, and that it has some kind of beak.

All in all,~here seerns to be a bundle of attributes which rep-resent important aspects of 'birdiness'. These attributes tend tocorrelate in nature, i.e. they appear together. A crea tu re that haswings and feathers is more likely to be able to By than one that hasa fur and four legs. Types of birds qualifying for these attributes

Attributes, good birds and bad birds: an example

Our starting point is a list of attributes collected for >ROBIN<,which was rated best example of the category BIRD in Rosch(1975; see Figure 1.3). The idea is that this list is the closest approx-imation available of the 'prototypical' attribute list for BIRD and thatit would be very suitable for comparisons with lesser categorymembers:

(3) 'has two wings and two legs'(4) 'has feathers'

(1) 'lays eggs'(2) 'has a beak'

24 An Introduction to Cognitive Linguistics

have a particularly prominent position in the category. Intermediateand bad examples of the category BIRD differ from these proto-typical examples in two ways: either they deviate to a moderatedegree with regard to one or more attributes (think of the parrot'slegs and tail) or some attributes are missing altogether (e.g. ostrichescannot fly).

This distribution is illustrated in matrix form in Figure 1.7,where a plus sign stands for an attribute which is present, a minussign for a missing attribute, and a plus/minus combination for agreater or lesser deviation from the expected form of the attribute.The matrix thus reflects the fact that a yes/no representation ofattributes (which would correspond to the dassical view) cannotadequately render the attributes ofbirds and has to be modified toindude intermediate judgements.

Though this may not be obvious at first glance, 'deviant' and'missing' attributes can be seen as different problems and have beenaddressed in different ways in cognitive research. Since the absenceof attributes (as in the case of >OSTRICH<) seems to be the moreserious problem, it will be tackled first, while the deviant attributes(as observed with >PARROT<) will be taken up later.

Prototypes and categories 25

there is no lack of attributes to describe >OSTRICH<, and whilesorne of these attributes ('is very tall', 'runs very fast') seem incom-patible with our idea of a bird, some are less so. Take the attribute'long neck' and you will find that this applies to birds as far apartfrom ostriches as are flamingoes and storks (though not to robins).Or take the attribute 'large soft feathers' and it may remind you ofswans, while the alternative version offered by other dictionariesfor this attribute, i.e. 'decorative feathers', suggests links with pea-cocks and perhaps even with parrots (though not with sparrows).

However, in other categories the items are even more dis-similar, as the philosopher Ludwig Wittgenstein showed in hismuch-quoted passagé about the category GAME:

Category members

Attributes >ROBlN< > SPARROW < >OOVE< >PARROT< >OSTRICH<

lays eggs + + + + +beak + + + + +two wings &two legs + + + + +/-feathers + + + + +small &lightweight + + +/- +/-can Ay + + + +chirps/sings + + + +/-thin/short legs + + + +/-short tail + + + +/-red breast +

Consider for examp1e the procecdings that we call 'games'. I mean board-games, card-games, ball-games, Olympic games, and so on. What iscornrnon to them all ? - Don't say: 'There must be something cornrnon,or they wou1d not be called "garnes" , - but 100k and see whether thereis anything common to all. - For if you 100k at them you will not seesomething that is common to al1, but similarities, re1ationships, and awho1e series of them at that. To repeat: dori't think, but 100k ! - Forexamp1e at board-games, with their multifarious relationships. Now passto card-games; here you find many correspondences with the frrst group,but many cornrnon features drop out, and others appear. When we passnext to ball-games, much that is cornrnon is retained, but much is lost, -Are they all 'amusing'? Compare chess with noughts and crosses. Or isthere always winning and losing, or competition between p1ayers?Thinkof patience. In ball-games there is winning and losing; but when a childthrows his ball at the wall and catches it again, this feature has disappeared.Look at the parts p1ayed by skill and 1uck; and at the difference betweenskill in chess and skill in tennis. Think now of games like ring-a-ring-a-roses; here is the e1ement of amusement, but how many othercharacteristic features have disappeared! And we can go through themany, many other groups of games in the same way; we see how simi-1arities crop up and disappear.

The principie of family resemblances

The paradox in the case of the ostrich is that we are quite preparedto call it a bird though it shares only few attributes with prototypicalbirds like the robin. Yet as the above dictionary definition suggests,

(Wittgenstein 1958: 66f)

Figure 1.7 Goodness-of-examp1e and distribution of attributes in thecategory BIRD

Wittgenstein's condusion was that games are connected by anetwork of overlapping similarities, which he called farnilyresemblances. On a somewhat more abstract level, the principleof farnily resemblances has been defined as

a set of items of the form AB, BC, CD, DE. That is, each item has at1east one, and probab1y several, elements in cornrnon with one or moreother items, but no, or few, e1ements are common to all items.

(Rosch and Mervis 1975: 575)

26 An Introduction to Cognitive Linguistics

Wittgenstein's explanation and the definition supplied byRosch and Mervis both pursue the same aim: to show that theprinciple of family resemblances opens up an alternative to the clas-sical view that attributes must be cornmon to all category members,that they must be 'category-wide'. This could be most impressivelydemonstrated with the analysis of 'superordinate categories' likeGAME (Wittgenstein's example) and FURNITURE, VEHICLE, FRUIT,etc., which were investigated by Rosch and Mervis (1975). As wewill see in section 2.2 of Chapter 2, it is not surprising that super-ordinate categories largely depend on family resemblances.

The picture changes when we consider categories like BIRDor other more concrete categories like CAR, TRUCK, AIRPLANE,CHAIR, TABLEand LAMP.As we observed in the case ofBIRD, evenvery bad examples of the category like >OSTRICH< (and we mightadd >PENGUIN<) have some important attributes in cornmon withall the other category members. And, of course, the good examplessuch as >ROBIN< and >SPARROW< all share whole bundles ofattributes, so that only a few of their attributes rely on the familyresemblance principle. This situation is illustrated in Figure 1.8,where attributes are indicated by lines.

Just as with birds, members of categories like CAR and CHAIR(i.e. different types of cars and chairs) share many more attributesthan the members of the superordinate categories VEHICLE andFURNITURE (i.e. different types of vehicles and different items offurniture). Here section 2.1 of Chapter 2 will provide an explana-tion.

Yet however much the significance of family resemblancesmay vary for individual categories, this does not affect the explan-atory potential of the underlying principle. What is decisive is thatfamily resemblances can explain why attributes contribute to theinternal structure of the category even if they are not cornmon toall category members, i.e. if they are not essential features accord-ing to the classical view. An ostrich is a bird not only because ithas feathers and lays eggs, like a robin (indicated by bold lines inFigure 1.8); category membership is also supported by the fact thatthe ostrich has a long neck like a flamingo, and decorative featherslike a peacock (see thin lines in Figure 1.8). In defining the posi-tion of a category member in its category, we are thus justified inconsidering any sensible attribute proposed for this item. This isthe theoretical background of the attribute listing experimentscarried out by Rosch and her associates and the typicality ratingsthat were based on them.

Prototypes and categories 27

>DOVE<>SWALLOWjXCK<~ a-f <,>HEN<>SPARROW< a-e __a-h ~

>BUDGERIGAR<a-e >ROBIN< a-e >GOOSE<--- -

>DU~J\k

>SWAN<

Seleetedeategory-wideattributes(a) layseggs(b) hasa beak(e) hastwo wingsandtwo legs(d) hasfeathers

Seleetedfamilyresemblaneeattributes(e) can fly(f) issmallandlightweight(g) ehirps/sings(h) legsare thin/short(i) keptin a eage(j) rearedforthe useof itsmeat,eggsandfeathers(le) haslongneek(1) hasdeeorativefeathers(m) hasexotieeolours

Figure 1.8 Selected comrnon attributes and family resemblances of thecategory BIRD

Attribute listing and attribute-based typicality ratings

Given the ease with which we seem to be able to call up the attri-butes for familiar objects and organisms, attributes can be collectedin a fairly simple test procedure which can be easily administeredto a large number of subjects. In the attribute-listing experimentsconducted by Rosch and Mervis (1975), each of the subjects (400American psychology students) was given six sheets of paper withthe test item written on the top of the page. The subjects had aminute and a half to write down all the attributes that they couldthink of To eliminate answers that were obviously false or wronglyattributed to an item or toa general in meaning, the attribute listswere checked by two judges.

The test items used were selected from the lists obtained inthe goodness-of-example ratings described in Figure 1.3 and

28 An Introduction to Cognitive Linguistics

consisted of sets of 20 graded category members, one set foreach of the categories FURNITURE, VEHICLE, FRUIT, WEAPON,VEGETABLEand CLOTHING. Altogether, 120 iterns from >CHAIR <to >TELEPHONE< (for FURNITURE), >CAR< to >ELEVATOR< (forVEHICLE) and from >ORANGE< to >OLlVE< (for FRUIT) weretested.

The experiment had two airns: to demonstrate the notion offamily resemblance (see above) and, more important for Rosch andMervis, to supply attribute-based typicality ratings (this neutral termseems preferable to Rosch and Mervis's own term 'measure offamily resemblance'). These typicality ratings could then be usedto verify the earlier goodness-of-example ratings.

How were the attribute-based ratings calculated? Leavingaside mathematical details, two stages can be distinguished. First,the attributes were 'weighted', that is, it was established for howmany of the 20 tested category members each attribute hadbeen listed. The top score of20 was given if an attribute was sharedby all category members ('means of transport' in the case ofvehicles). An attribute listed for only one category member (thinkof 'installed in buildings', which would only fit >LlFT< or>ELEVATOR <) received the score 1; attributes applying to severalbut not all category members were assigned intermediate scores.The result was a list of weighted attributes.

In the second stage the weights of all the attributes listed foreach category member were added up and rank-ordered (the actualprocedure was somewhat more complexj.I'' Category memberswith many shared and therefore highly weighted attributes, suchas >ORANGE< or > APPLE< in the FRUIT category, achieved highoverall ratings; category members like >OLlVE<, which shared onlyfew, if any, attributes with other category members, scored low.When this attribute-based rank order was compared with thegoodness-of-example ratings, the two types of ratings showed ahigh degree of correlation.

This correlation could be used to support the notion of proto-type categories in two ways. On the one hand, the hypothesis thatthese categories consist of good and bad members was no longersolely dependent on the intuitive judgements of the goodness-of-exarnple ratings, but could now be related to a large range ofattributes. On the other hand, the notion of good and badexamples could be used to explain why attributes are so unevenlydistributed among category members. While good examples havemany attributes in cornmon with other members of the same

Prototypes and categories 29

category, bad or marginal exarnples share only few attributes withmembers of the same category.

Yet as we know from the dictionary entries discussed at thebeginning of this section, the intra-categoriallinks of attributes rep--resent only one side of the coin. Equally important is the questionof distinctive attributes, i.e. whether the members of a category,both good and bad examples, share attributes with membersof neighbouring categories. This was tested in a further attribute-listing experiment in which good and bad exarnples of the categoryCAR were contrasted with members of the categories TRUCK, BUSand MOTORCYCLE. As it turned out, the overiap of attributes issmallest between the good examples of the different categories, butis much larger in the case of bad or marginal examples, In otherwords, a prototypical car, saya saloon, has fewer attributes in com-mon with a prototypical truck than an estate car does; the saloonalso shares fewer attributes with a prototypical motorcycle than athree-wheeler does.

Looking now at both sides of the coin we can surnmarize theattribute structure of prototype categories as fol1ows:

(1) Prototypical members of cognitive categories have the largestnumber of attributes in cornmon with other members of thecategory and the smallest number of attributes which alsooccur with members of neighbouring categories. This meansthat in terrns of attributes, prototypical members are maxim-ally distinct from the prototypical members of other categories.

(2) Bad exarnples (or marginal category members) share only asmall number of attributes with other members of their cate-gory, but have several attributes which belong to othercategories as well, which is, of course, just another way of say-ing that category boundaries are fuzzy.

Attributes and dimensions

When we return to our initial matrix ofbird attributes at this point,we find that Rosch and Mervis have indeed solved the problem of'rnissing' attributes as observed with >OSTRICH< and other badexamples of the bird category. This still leaves us with the 'deviant'cases, that is category members whose attributes do not complywith the expected norm. Most of these 'deviant cases' involve attri-butes related to dimensions, such as 'size' and 'weight', 'length(oftail)' and 'thickness (oflegs)'.

30 An Introduction to Cognitive Linguistics

Figure 1.9 A selection of the drawings of cup-like objects used byLabov (1973: 354, no. 5 reconstructed)

This takes us back to Labov's cups and bowls, which werealready discussed in the context of fuzzy category boundaries insection 1.1. In choosing cups and cuplike containers for rus experi-ments Labov could rely on 'width', 'height/depth' and 'shape', i.e.on generally accepted 'logical' properties of the dimension type,and did not have to expose hirnself to the vagaries of empiricalattribute listing. In this respect he is more in line with traditionallinguists than Rosch and Mervis are. The dimensions had theadvantage that they could be varied under controlled conditions,and these variations could be easily and unequivocally representedin line drawings. This was illustrated in Figure 1.4, which is hererepeated for convenience as Figure 1.9.

In this figure the variation of the dimensions is based on the'prototypical' cup no. 1. In the top row the height (or depth) ofthe vessels is kept constant while the width is extended systematic-ally from vessel to vessel. Conversely, in the second row the widthof vessels is identical, but from left to right the vessels becomedeeper and deeper. Other scalar variations used by Labov (and notillustrated in Figure 1.9) concerned the shape of the vessels andinvolved cylinders, truncated cones, etc. In addition, he used attri-butes like 'context' or 'function' ('neutral! coffee/food/ftowercontext'), 'material' (glass, china) and 'presence/absence ofhandle(s)', all of them 'non-scalar' attributes at first sight.

To bring these diverse attributes in line with dimensions,Labov defined the specific width/height ratios of rus test vessels asdiscrete values on the width and height dimensions. In the sameway the attributes involving context, material and handles wereinterpreted as values on a dimension: the neutral, coffee, food andftower contexts as values on the 'context' dimension, glass andchina as values on the 'material' dimension; presence and absenceofhandles as two values on the 'handle' dimension.

Prototypes and categories 31

How did these attribute values show up in the actualtest results, i.e. the use of cup or bowl or mug for certain vessels?According to Labov, the use of these words reftected the 'weighted'acceptability judgements of test subjects. These judgements wereweighted in the sense that the different values of the attributesinvolved led the subjects to choose different names for the test ves-sels. If the 'prototypical' cup (vessel no. 1) was called cup by all testsubjects, this judgement could be related to the fact that vesselno. 1 represented the favourite width/height ratio in a neutral con-texto Vessel no. 2 was still overwhelrningly called cup because itrepresented a very similar width/height ratio. In contrast, vessel no.4 was assessed as a borderline case of cup because its width/heightratio differed markedly from the favoured value. If the use of cup wasextended by the test subjects in a coffee context and reduced in afood context, this was due to the interaction of the width/heightratio with the respective values on the 'context' dimension.

While such three-variable constellations could still be docu-mented in consistency profiles (as shown in Figure 1.5 and 1.6),the more complex interactions could only be mastered by mathe-matical formulas and probability calculation which are beyond thescope of this introduction (see 7 in section 1.1).

Nevertheless, the airns and the methods employed by Labovshould have be come clear. Starting out from attributes whichcould, at least theoretically, be regarded as values on dimensions,he overcomes the lirnitations implicit in the discrete attributes ofthe logical view, and this is where he seems to go further thanRosch and Mervis. The scalar values are then related to each otherin a weighting process, and here rus method comes closest toRosch and Mervis's procedure.

Internal category structure and gestalt

When we compared attribute-based typicality ratings with thegoodness-of-example ratings (i.e. direct typicality ratings), we sug-gested that the latter are based on intuition. The question is,however, what this intuition rnight be derived from. Does it per-haps rely on some internal attribute-based rating? From a 'naive'point of view, this seerns unlikely. When we encounter an animalwe will hardly begin categorizing by evaluating specific attributes,unless we are struck by a very salient one like the strip es of a zebraor the trunk of an elephant. With most organisrns and concreteobjects, especially those that are familiar to us, we seem to proceed

32 An Introduction to Cognitive Linguistics

in a different way. We simply take in an overall picture of thewhole and use it for a first assessment of its goodness. The con-sideration of specific attributes can then be left untillater.

The problem with this quite plausible conception is that itseems impossible to investigate it in controlled test situations or, toput it more pointedly, to exclude experimentally that people cate-gorize and evaluate the goodness of items on an attribute-by-attribute basis. To return once more to Labov's naming tests, it doesindeed make sense to assume that informants' judgements are basedon the overall impression of a cup or bowl rather than on an internalcomputation of the width/height ratio, material and context, butthis alternative cannot be ruled out. For the time being, we have tobe content with results of more informal test interviews.

A series of such interviews was conducted in the context ofthe categorization of houses (Schmid 1993: 121ff). The startingpoint was Labov's method of eliciting category names for the draw-ings of objects, which was applied to the buildings shown in Figure1.10. The aim ofthe interview part ofthe test was to collect infor-mation about how the categorization process was carried out andhow it was experienced by the informants.

After a first glance at the drawings (which were presented oneby one with other pictures interspersed) the informants named thebuildings. At this stage, pictures (a) and (b) were quite readily cate-gorized as cottages, while picture (c) was assigned to adjacentcategories which were labelled house in the country or villa. Just aswith Labov's cups and bowls there was no total agreement amongthe informants. While 10 of the 12 informants did not call picture(c) cottage, two were prepared to do so. When asked to give thereasons for their categorizing decision, the informants explained theyhad judged on the basis of a general impression of the drawings. Thiscan be taken as an indication that an internal goodness-of-examplescale was already established on the basis of the first overall impres-sion. Only on second thoughts were the informants able to pinpointcertain individual properties that might have led them to select thename they did. The choice of cottage seems to have been influencedby the material ('made of stone') and the simple but sturdy construc-tion in the case of picture (a), and by the thatched roof for picture(b). The deviant width/height relation (,too big') and the numer-ous extensions ofbuilding (c) may have helped to rule out cottage asan appropriate name for the majority of the informants.

As far as these interviews go, they support the initial assump-tion that categorization and goodness ratings may indeed involve

Prototypes and categories 33

(a) (b)

Figure 1.10 Examples of houses used for a categorization task (Schmid1993: 15lf)

two stages: the perception of an object as a whole as the first step(the so-called holistic perception), and a kind of decompositionof the perceived whole into individual properties or attributes as asecond (optional) step.l '

The idea of 'perceived whole' comes close to the notion ofgestalt as advocated by gestalt psychologists, so it may be helpfulto look at some of their findings. Their major claim is that gestaltperception can be traced back to 'gestalt laws of perceptual organ-ization', or 'gestalt principles', which are usually demonstrated withline drawings and dot pattems.l/ The most important of these prin-cipIes are:

o 'principle of proximity': individual elements with a smalldistance between them will be perceived as being somehowrelated to each other.

o 'principle of similarity': individual elements that are similartend to be perceived as one common segmento

o 'principle of closure': perceptual organization tends to beanchored in closed figures.

o 'principle of continuation': elements will be perceived aswholes if they only have few interruptions.

34 An Introduction to Cognitive Linguistics

The more a configuration of individual elements adheres to theseprinciples, the more it will tend towards a clear-cut and cogentorganization (called Prdgnanz: by the gestalt psychologists), whichlends itself to gestalt perception. Examples which show a highdegree of Prágnanz: are called 'good gestalts' or 'good forms'.Among them are circle, square and equilateral triangle, which wealready encountered in our discussion of shapes in section 1.1,where they appeared as natural prototypes of shape categories.

However, shape is just one aspect of the gestalt of organismsand objects, although such an important one that it is sometimes,inaccurately, treated as equivalent with gestalt. The question iswhether the gestalt principles are valid not only for dot patternsand geometrical shapes, but can also be applied to the much morecomplex configurations of organisms and objects.

As it seems, a central rol e in providing an object with agestalt is played by its constituent parts. Consider once more thecottages (a) and (b) in Figure 1.10. lt is obvious that they consistof walls, windows, a door, a roof and a chirnney. Yet when welook at these drawings for the first time, the cottages are not visu-ally 'deconstructed' into these parts, but are perceived as an integralwhole. This holistic visual perception is possible beca use the partsare organized according to the gestalt principles of proxirnity (allthe parts are close together), of sirnilarity (identical windows), ofclosure (all the parts are included in one overall outline) and con-tinuation (all the lines are uninterrupted).

Sirnilarly, if you look at some everyday object that is in yourview at the moment, maybe a chair, a table or a bookshelf, youwill realize that the salient parts make an important contributionto the whole without at first being noticed as individual parts.Other perceptual aspects such as the overall proportions, thematerial and the colour of objects interact with the overall shapeand the parts to complete the holistic impression of a gestalt.

Having established that objects are perceived as integralwholes, we can go on to ask ourselves what makes one gestaltmore prototypical for a category than another. Here we enter newground, because apart from the good forms among geometricalshapes, this question was not really the concern of gestalt psycho-logy. We would like to argue that it is again the parts of an objectthat play an important role in establishing a prototypical gestalt.Parts do not only contribute to the overall shape of an object, butare also related to its function, which for most artefacts is theraison d'étre of the object. Likewise, the existence of most parts of

Prototypes and categories 35

(a)

(b) (e) (d)

Figure 1.11 A prototypical chair gestalt and other chairs

an object is motivated by the particular purpose the objectserves.13

Consider a chair, for example. The functionally relevant partsof a chair are the legs, the seat and the back. Our assumption nowis that the prototypical gestalt of a chair relies predorninantly on thepresence of these three functional parts in optimally functional pro-portions. Looking at the drawings of chairs shown in Figure 1.11,the prototypical chair gestalt is best represented by picture (a).

An analogous explanation can be attempted for the gestalts oforganisms such as birds if we understand the 'function' of a bird tomean the enactment of a certain forro of organic life. This proto-typical mode of life would be characterized by the ability to liveon trees and shrubs, to By, and to pick nuts, seeds, worms, etc., forfood. Assurning a functional design of the parts, we would expectfrom a prototype gestalt that parts like the beak, the legs, andespecially the wings, have the appropriate form: the legs shouldhave claws which allow the bird to cling to branches, the beakshould be protrucling and sharp to facilitate food picking, wingsshould be shaped aerodynarnically. Even more important, the partswould have to have the right proportions (large wings in com-parison to the remaining part of the body, etc.) This may read likea replica of the attributes listed in the matrix at the beginning ofthis section (see Figure 1.7). The decisive clifference is that these

/

36 An Introduction to Cognitive Linguistics

functionally balanced parts are all integrated into one gestalt, andare perceived as a whole.

All this might suggest that for concrete objects as well as fororganisms the prototype gestalt should be conceived as a kind ofreduction to the relevant and mainly visual essentials, though per-ceived as a whole. Such a notíon is supported by the experienceof how easy it is to grasp and categorize reduced illustrations, ashave been increasingly used in dictionaries over the last few years.l"(The case of pictograms, as used in traffic signs, is more complexbeca use here the drawings are used as a vehicle for instructions andwill therefore be neglected.).

We would hold the position that the effectiveness of reduceddrawings is not a general phenomenon even with concrete objectsand organisms, but depends on the kind of lexical category to beillustrated. Just look at the drawings of a bungalow and a cottagein Figure 1.12, which are both taken from the same dictionary.The bungalow is reduced to the functional parts (one-storey build-ing with walIs, roof, windows, door), and can still be readilyidentífied. In contrast, the iilustration of the cottage contains awealth of detail, and this seerns to be necessary for easy recogni-tion, so it is in fact more easily categorized than some of the plainerdrawings of cottages in Figure 1.10.

From this we deduce that the prototype gestalt of a cottagecomprises more than the complete and well-proportioned parts ofa house, important as ail this may be. The function of a cottageincludes notions like creating warmth and cosiness and beingembedded in natural rural surroundings. In other words, all sortsof emotional and attitudinal properties are involved. Similarly it isdifficult to imagine a line drawing of an English pub (and, incid-entally, none ofthe major dictionaries provides one); the wealth ofgestalt properties applying to a pub is simply too great.

bungalow

Figure 1.12 Bungalow and cottage: examples of reduced and 'rich'dictionary illustrations (from LDOCE)

Prototypes and categories 37

Or take another simple case, the teddy bear. To approximatethe prototype gestalt of a teddy bear it is certainly not enough togive the outline drawing of a teddy that coincides with its naturalmodel, the brown bear. What is also needed is an indication of itssoftness (something that takes us beyond the visual to the tactileproperties of the gestalt), and, perhaps most important, the teddyshould be hugged by a child to round off the general impressionof its function.

This shows that even in the domain of concrete objects weare not just confronted with a single kind of gestalt prototype (the'reduced type') but have to take into account that 'Certain lexicalcategories require richer visual representations, which in turn sug-gest richer underlying gestalt protorypes.: The situation becomeseven more complex when we consider that for many lexicalcategories - those loosely called 'abstract' categories - gestalt per-ception is largely excluded.

Summing up at this point, our discussion of gestalt has shownmainly two things:

(1) Gestalt perception seems indeed to play an important part incategorization and goodness ratings. For the categories dis-cussed it seems to be as essential as the possibility of studyingattributes and farnily resemblances and computing attribute-based typicality ratings.

(2) The role of gestalt in the categorization of objects and organ-isms need not be completely left to intuition, it can be studiedby making selective use of the principIes of gestalt psychologyand by considering additional aspects like parts and function.

Any research into gestalt properties, and even the prelirninarysketch attempted here, is intimately bound up with questions ofmental representation and processing, just as the notions of goodand bad examples, of attributes, and of prototypicality in generalcannot be isolated from their cognitive background. So it seemsappropriate to conclude this section with a few more generalremarks about the cognitive status of categories and of the notionsinvolved in categorization.

.The cognitive'status of categories, prototypes,attributes and gestalt

For the linguist, categorization is an important issue beca use itunderlies the use of words and the use oflanguage in general. Since

• j

Prototypes and categories 39

As this short survey of the logical and the prototype paradigmshas shown, it do es indeed help to take the strictly cognitive view,rhe view that categorization is something that underlies the men-tal processes oflanguage comprehension and language production.This view provides us with a vantage point from which we cannow delve deeper and examine the main elements of the proto-type hypothesis: the notions of prototype, category membershipand typicality, attributes, farnily resemblances and gestalt.

38 An Introduction to Cognitive Linguistics

producing and understanding language undoubtedly involve cog-nitive processes, categorization is necessari1y something that takesplace in our minds, and the categories resulting from it can beunderstood as mental concepts stored in our mind. Taken togetherthey make up what has been called 'mental lexicon' .15

Unfortunately, we do not have direct access to cognitive pheno-mena, so everything that is said about the categories of the mentallexicon can only have the status of a more or less well-foundedhypothesis. Such a hypothesis can be supported by philosophicalargument, by physiological research into the human sensoryapparatus and by experimental evidence based on linguistic andother human behaviour.

To avoid misunderstandings, two things should be kept inmind. First, the number of hypotheses about cognitive categoriza-tion is theoretically unlimited; it is definitely not restricted to thetwo paradigms which will be discussed in this section. Second, thepostulated category paradigms need not apply indiscriminately tothe whole domain of human categorization, but may be restrictedto certain areas and perspectives.

Let us start with the 'classical' or 'logical model' of categor-ization, a hypothesis with a very long tradition. This hypothesisclairns that categories come as homogeneous units with clear-cutborderlines and that all members are characterized by a limitednumber of essential features (a set of necessary and sufficient condi-tions, as discussed above). This beautifully simple model is oftenaccompanied by the philosophical speculation that this type ofcategory mirrors, or is even predetermined by, the constitution ofthe organisms and objects in the 'real' world. The problem is thatthis logical hypothesis is not in accordance with the evidence col-lected by physiologists and psychologists.

In contrast with the classical model, the experiential proto-type hypothesis of categorization claims that categories are nothomogeneous, but have a prototype, good and bad members, andhave fuzzy boundaries. Category members do not all share thesame discrete attributes, but may be linked by family resernb-lances. In the case of colours and shapes, prototype theory issupported by both physiological and psychological evidence. Assuggested in section 1.1, colour categories and focal colours seemto be based on the nature of the human perceptual apparatus. Theirprototype structure was also confirmed by psychological tests. Inthe case of organisrns and concrete objects only this second type ofevidence is available.

Prototype: Basically, there are two ways to understand the notion ofprototype. It can be deduced from categorization experiments. Forinstance, some rnernbers of a category first come to mind in asso-ciation experiments and are recognized more rapidly as categorymembers in verification tasks. If one takes these members as pro-totypes of the respective categories, this leads to definitions like'best example of a category', 'salient examples', 'clearest cases ofcategory membership', 'rnost representative of things included ina class' or 'central and typical members' (see Rosch 1978; Lakoff1986; Brown 1990; Tversky 1990).

But this is not the conception of prototype that we wouldadvocate. Instead we would claim that if one takes the cognitiveview of categories seriously, one is justified in defming the proto-type as a mental representation, as some sort of cognitive referencepoint. Thinking of prototypes in this genuinely cognitive way stillleaves open the nature of the mental representation, so thatdefinitions may range from the more concrete notion of 'image'or 'scherna' to the more abstract 'representation of a category' or'ideal' according to the categories to which they are applied(defmitions from Rosch and Mervis 1975; Coleman and Kay 1981;Lakoff 1986).

Category membershíp and typícalíty: Unlike the homogeneous cate-gories postulated by the logical hypothesis, cognitive prototypecategories always consist of good and bad members and include mar-ginal examples whose category membership is doubtful.l? This notonly applies to attested cases of prototype categories, such as colourcategories or CUP, BOWL, MUG and CHAlR, but also to categories likeBlRD, where category membership seerns to be safely based on dis-crete attributes such as 'taying eggs'. But do ordinary language usershave the encyclopaedic knowledge to decide whether a penguin layseggs or not? If they do not, the issue whether the penguin is a birdwill remain undecided, and >PENGUIN< will be a doubtful memberof the prototype category BIRD.

40' An Introduction to Cognitive Linguistics

Even for prime cases of apparendy discrete homogeneous cat-egories, such as ODD NUMBER, SQUARE and kinship categories(MOTHER, UNCLE, etc.), a prototype structure cannot be completelyexcluded. As experiments have shown (Armstrong et al. 1983; Fehrand Russel 1984), informants do in fact distinguish between goodand bad examples of odd numbers and squares, and similar reactionscan be assumed for MOTHER (Lakoff 1987) and other kinship terms.

However, this only applies to 'everyday' categorization. In amathematical or scientific context the logical or classical viewcomes into its own. In such a context NUMBER, SQUARE or kin-ship categories can be established as clear-cut and homogeneouscategories by an act of definition. In other words, the classicalparadigm of categorization has a wide field of application whereverthere is a need for precise and rigid definitions as in the domain ofscientific categorization or in the legal field. And there is no reasonwhy the discrete categories of science and the everyday prototypecategories should not coexist in the mentallexicon and even influ-ence each other.

/

Attributes and Jamily resemblances: Attributes as col1ected from infor-mants by Rosch and others are statements which pro videinformation about the members of a category. In this sense attri-butes are part and parcel of the empirical investigation of categorystructures.

When the attribute lists for individual category members arecompared, this is assumed to reflect the similarity relations betweencategory members (Taylor 1995). Such sirnilarity relations mayencompass al1category members ('category-wide attributes') or theymay establish links onIy between some of the members. In the lat-ter case, category coherence is produced by family resemblances.

Since similarity relations between good and bad members of acategory are part of the hypothesis of prototype structure, this impliesthat they are also part of the mental representation of a category. Thisis not necessarily so for the empirical1y col1ected attributes which areelicited from informants and which may be fragmentary and over-lapping. So attributes are best considered as a descriptive tool andnot as part of the mental representation of the category. Such acautious position is even more advisable when attributes are treatedas values on dimensions. For such dimensions as 'size', 'width' and'shape' may be imposed on the attributes by the logically rnindedresearcher and are not necessarily a reflection of our natural way ofthinking about the objects in the world around uso

Prototypes and categories 41

Gestalt: As original1y conceived by gestalt psychologists, the notionof gestalt was intended as an explanation of holistic perception. Inthe last section we suggested a link between gestalt and the notionof prototype categories. If a gestalt is organized according to thegestalt principles and includes the functional parts of an item infunctional1y balanced proportions, it may be regarded as a 'proto-type gestalt'. This ties in with the definition of prototype as an'image', which was quoted above. In fact, in the case of organismsand concrete objects where visual perception seems to be impor-tant, the prototype gestalt contributes considerably to the ability ofthe prototype to function as a model or cognitive reference point.

This is as far as we think we should take the discussion of thepsychological issues at this point. To complement this psycho-logical view, the fol1owing section willlook at some sociological,or more broadly, cultural factors that influence the formation ofprototypes and prototype categories.

Exercises

1. Col1ect attributes for the fol1owing categories and try to dis-tinguish between objective properties and subjectiveassociations:

MAN, WOMAN, BOY, GIRL;MANSION, PALACE, COTTAGE, CASTLE;BICYCLE, MOTORBlKE, CAR, VAN, LORRY;jEANS, LEGGINGS, TUXEDO, TAILCOAT, MINISKIRT;

2. Look at the picture of the 'Smith brothers' devised byArmstrong et al. (1983: 269).

42 An Introduction to Cognitive Linguistics

In what way is the principle of family resemblances illustratedby this drawing? Which attributes are shared by all thebrothers?

3. Have a look at how Wierzbicka (1985: 19-36) explains thedifference between cUP and MUG and discuss whether youfind her arguments plausible. Wierzbicka claims that the dif-ference in meaning boils down to the difference in theintended use of the respective objects; apply this idea to thecategories VASE, BOWL and PLATE.

4. Discuss the attributes 'fun', 'no purpose other than the gameitself, 'uncertain outcome' and 'govemed by rules' as candi-dates for the status of category-wide attributes for the categoryGAME. Try to come up with other possibilities and discuss pos-sible counterexamples.

5. Pictograrns are more than simple line drawings. Look at sometraffic signs and some of the signs in railway stations, busterminals or airports and explain whether they make use ofgestalt perception. What additional information do theyconvey?

6. Ask a friend or two to draw simple pictures of a car, a bus, atelephone, a book, a bottle, a tree, a house and a church.Discuss to what extent the principies of gestalt psychology canbe applied to their drawings. Does the perspective or vantage-point from which the objects are viewed playa role?

7. Discuss the difference between the logical view of categoriza-tion and the prototype model with reference to legal categorieslike MURDER, MANSLAUGHTER; VANDALISM, LOOTING; ASSAULT,ASSAULT ANO BATTERY; NEGLIGENCE, GROSS NEGLIGENCE.What are the problerns a judge faces when he has to deal withunclear cases?

8. In the context of the acquisition of the meaning of wordsAitchison (1994: 174f) writes about a child's use of the wordqua ('quack') as discussed by the Russian psychologistVygotsky:

The child began with qua as a duck on a pond. Then the liquid ele-rnent caught the youngster's attention and the word was generalizedto a cup of rnilk. But the duck had not been forgottcn, and this sur-faced in qua used to refer to a coin with an eagle on it. But then thechild appeared to ignore the bird-like portion of the rneaning andfocus only on the roundness of the coin, so reapplied the word quato a teddy-bear's eye.

Prototypes and categories 43

Discuss how the strange assembly collected by the child in thecategory QUA can be explained with the principle of familyresemblances.

1.3 Context-dependence and cultural models

The prototypes cif cognitive categories are not fixed, but may changewhen a particular context is introduced, and the same is true forcategory boundaries, More general/y, the whole internal structure o] acategory seems to depend on the context and, in a wider sense, on oursocial and cultural knowledge, which is thought to be organized incognitive and cultural models.

Reading the sentence He opened the door toface a pretty young womanwith a dog in her arms, what kind of dog would first spring to yourrnind? Would it be an Alsatian or a collie, which would presum-ably turn out to be prototypical dogs in a goodness-of-examplerating? Or would you not rather think of a pekinese or some otherkind of small lapdog? The chances are that you would. Thoughthis may look like a rather trivial example, it has far-reachingimplications for the theory of prototypes. What it suggests is thatprototypes are not after all the fixed reference points for cognitivecategories that we have assumed them to be, but that they are liableto keep shifting as the context changes.

The context-dependence of prototypes and of the .whole internal category structure

Let us first expand the lapdog example and, by way of a littIeexperiment, compare it to other examples involving references todogs. Read the following four sentences one by one, pausing fora second after each example to check what kind of dog is suggestedto you:

(1) The hunter took his gun, left the lodge and called his dogo(2) Right from the start of the race the dogs began chasing the

rabbit.(3) She took her dog to the salon to have its curls reset.(4) The policemen lined up with the dogs to face the rioters.

We can safely assume that for each of the four sentences you willhave formed a different image of the kind of dog that is denoted.In a hunting context like example (1), the most likely dog would

44 An Introduction to Cognitive Linguistics

probably be some kind of retriever; in the dog racing =v= ofexample (2), it would certainly be a greyhound that would firstcome to mind. In examples (3) and (4), you will presumably haveimagined a poodle and an Alsatian respectively. These examplessuggest that what turns out the most likely member of a certaincategory depends on the contexto Since we may expect the proto-type to be our first choice, the result of our little experimentindicates that, depending on the context, the prototype shifts. The'context-dependent' categories which are thus evoked may becompletely different from the non-contextualized prototypeselicited in goodness-of-example experiments.

However, more than a shift of prototypes is at stake in theseexamples. Let us assume that in a goodness-of-example test>ALSATIAN< would be rated as a prototypical dog, with>GREYHOUND<, >SETTER < and >POODLE< beingjust slightly lessprototypical. >PEKINESE< would presumably be a rather peri-pheral dog because of its small size, flat face and somewhat'undoggy' behaviour. Taking this category structure as a starting-point, it seems clear that the sentences above do more than justshift the prototype. In some cases the category structure of thecontext-dependent category is much leaner than that of the non-contextualized category. For example in sentences (2) and (3) thecontext-dependent prototypes are >GREYHOUND< and >POODLE<respectively; virtually all other types of dogs are so unlikely that,for all practical purposes, they are highly peripheral members of thecontext-dependent category.

In contrast, in sentence (1) the internal category structure ofDOG is retained to a larger extent. One could certainly imagine anAlsatian or a setter being used as a hunting-dog. Nevertheless theyare clearly less typical members of the context-dependent cate-gory (HUNTING-)DOG than retrievers. Thus the principle of differ-ent degrees of goodness-of-example is still valid in this case, andthe same is true for examples (3) and (4). The main point, how-ever, is that the context-dependent category structure is differentfrom the structure that was obtained in non-contextualized good-ness-of-example ratings. Altogether, it seems that the context notonIy determines the choice of the category prototype, but that italso leads to an adjustment of the position of other category mem-bers. How canthis be explainedrl"

One way is to use attributes as an explanatory tool. In section1.2 we saw how weighted attributes can be employed to explainthe typicality structure of a category. When we view it in terms of

Prototypes and categories 45

attributes, context seerns to have a twofold effect: first, the contextcan change the weight of attributes that seem to be relevant for acertain category. Attributes of the category DOG that seem to bedecisive in the goodness-of-example ratings and attribute-basedtypicality ratings (e.g. 'barks', 'has four legs', 'wags tail when happy','likes to chase cats', etc.) apparently lose weight in specific contexts;second, the context can emphasize attributes that are not promin-ent and even introduce new attributes which would not bementioned at all in non-contextualized attribute-listing experiments.In the hunting-dog context, attributes like 'brings back the kill' or'points out the position of animals for shooting' increase in impor-tance. In the dog-racing context, 'has long, thin legs', 'can run fast','is enduring' and others become crucial. With the introduction ofnew attributes and the re-evaluation of the weights of existing onesthe attribute list for a member of a category changes completely.The result is that previously peripheral examples are equipped withlarge bundles of heavily weighted attributes and tumed into goodexamples or even prototypes, while well-established good examplesare reduced to the status of marginal mernbers.

Context, situation and cognitive models

Once we have acknowledged that the context can completelyreshuffie the positions of members within the category structure,we must ask ourselves what the context is. How can we grasp thisrather elusive notion, which is nevertheless one of the mostwidely used terms in linguistics?

'Context' has been defmed in many ways by scholars with dif-ferent backgrounds and various aims in mind. From a purelylinguistic point of view the context has been regarded as the lin-guistic material preceding and following a word or sentence.Language philosophers and pragmalinguists, most notably Searle(1979: 125), have defmed context as the set ofbackground assump-tions that are necessary for an utterance to be intelligible. Indiscourse-oriented approaches to language the context has beenre1ated to the situation in which an utterance is embedded.Originating in the work of the anthropologist Malinowski, theterm 'context' has been extended still further to include the so-called 'context of culture'. Malinowski had argued as early as 1923that both the 'context of situation' and the 'context of culture'were necessary for a proper understanding of an utterance or text(Halliday and Hasan 1989: 5ff).

46 An Introduction to Cognitive Linguistics

For cognitive linguists it is important that the notion of'con-text' should be considered a mental phenomenon. Thisrequirement is stressed by Langacker, for example, who defines hiscentral notion of domain 'as a context for the characterization of asemantic unit' (1987 a: 147). This very general interpretation of'context' will be discussed in Chapter 4 alongside other basic tenetsofLangacker's theory. In this section we will try to present a moretangible, but still cognitive, view of the notion of' context'.

To do this it will be helpful to distinguish 'context' from'situation' and a few other related terrns. Figure 1.13 illustrates oursuggestion for a meaningful terrninological distinction between thenotions 'context' and 'situation' with reference to the example sen-tence The boy was building a sandcastle with his bucket and his spade.The figure shows that we will treat the 'context' as belonging tothe field of mental phenomena, while the 'situation' refers to somestate of affairs in the 'real world'. (The quotation marks around realworld are necessary in view of the age-old debate among philos-ophers as to whether there is such a thing as a 'real world' at all;see 6 in section 1.1).

As Figure 1.13 indicates, we define the term situation as theinteraction between objects in the real world. In the exemplarysituation described by, the sentence above, the situation is made upby four objects, namely a boy, a sandcastle, a bucket and a spade,

The 'real' world The mind

object 1 concept/ category 1

t 'a young male cognitivehuman being' representationinteraction object 2 concept/ category 2

of interactionbetween betweenobjects

'a small model of acategories

~situation..... " ........ castle built of sand' context

'a boy 'building abuilds a object 3 concept/ category 3 sandcastle'sand-castle f) 'open container madewith of plastic or metal'a bucketand aspade' object 4 concept/ category 4

~'tool for diggingearth, sand, etc'

Figure 1.13 'Situation' and 'context' illustrated with reference to thesentence The boy was building a sandcastle unth. his bucket and his spade

Prototypes and categories 47

which interact through the activities of the boy. When the sen-tence is being processed by the hearer or reader, the words call upthe corresponding cognitive categories, or to put it more simply,the mental concept which we have of the objects in the real world.In addition, a cognitive representation of the interaction betweenthe concepts is formed, and it is for this cognitive representationthat we would like to reserve the term context.J'' In Figure 1.13,we have characterized the context for the example sentenceroughly as 'building a sandcastle'., Of course, this cognitive representation, or context, do es not

remain an isolated mental experience, but is irnmediately associ-ated in at least two ways with related knowledge stored inlong-term memory. On the one hand, context-specific knowledgeabout the categories involved is retrieved.l" This leads to the selec-tion of >PLASTIC BUCKET< as the most typical member of thecontext-dependent category BUCKET and of >CHILDREN'S SPADE<as the most typical member of the context-dependent categorySPADE.

On the other hand, the currently active context calls up othercontexts from long-term memory that are somehow related to it.In our example, experiences about other aspects of sand and sand-castles, besides those expressed in the sentence above, rnight beevoked. Related categories like WATER, HANDS, SHELLS, TURRETor MOAT, and frequent interactions between all these categories like'digging sand' or 'shaping turrets' are good candidates for such asso-ciated contexts.

It stands to reason that for all kinds of phenomena that wecome across in everyday life, we have experienced and stored alarge number of interrelated contexts. Cognitive categories are notjust dependent on the irnmediate context in which they areembedded, but also on this whole bundle of contexts that are asso-ciated with it. Therefore, it seems quite useful to have a termwhich covers all the stored cognitive representations that belongto a certain field. We will use the term cognitive model for theseknowledge bases (and indicate them typographically by usingunderlined small capitals); other related terms like 'frame' or 'script'will be taken up again in Chapter 5.20

Figure 1.14 shows one of the two cognitive models (ON THEBEACH and IN THE SANDPIT) in which the context 'building a sand-castle' could possibly be embedded. When we consider thisexample, two important properties of cognitive models becomeirnmediately apparent. First, as indicated by the reference to 'other

48 An Introduction to Cognitive Linguistics

contexts' in Figure 1.14, cognitive models are basically open-ended. A practical consequence of this property is that it is veryhard to describe the cognitive model of a domain and that descrip-tions of cognitive models are never exhaustive, but always highlyselective.

Second, just like the contexts that build the basis for cogni-tive models, cognitive models themselves are not isolated cognitiveentities, but interrelated. In our example, it can be seen that thecategories PEOPLE, SEA, SAND and others keep occurring in variouscontexts that make up the model ON THE BEACH. Consequently,the cognitive models of PEOPLE, SEA and SAND are closely relatedto the model ON THE BEACH. In line with recent theories of thernind which emphasize its so-called 'connectionist' architecture,one can argue that cognitive models combine to build networks.é!This tendency can probably best be explained using a visual repre-sentation. Figure 1.15 gives an idea of a network which consists ofvarious cognitive models that are interrelated through multipleconnections.

So far, two aspects of cognitive models have emerged fromthe exemplary representations that we have provided: their incom-pleteness and their tendency to build networks. There is a third,although not so obvious, property of cognitive models that should

context 2'sunbathing'

(pEOPLE, SUN, SAND, TOWEL,DECK CHAIR, SU GLASSES;

lie down, sleep)

context 1'swirnrning'

(pEOPLE, SEA,WAVES; wa1k,

swim, dive)

context 3'building a sandcastle'

(pEOPLE, SAND,BUCKET, SPADE,

dig, shape, build)

othercontexts

context 4'picnicking'

(pEOPLE, SANDWICHES,COKE, ICE-CREAM;

eat, drink)context 5'walking along the beach'

(pEOPLE, SAND, WAVES, SHELLS;walk, talk, look, pick up)

Figure 1.14 Schematic illustration of the cognitive model ON THE

BEACH (the major categories and the way they interact are indicated inbrackets)

Prototypes and categories 49

SUN ------- SAND ------- SEA

WEATHER -+--- BEACH -----+- });WE~E~KE~ND1>/,..c----t- FlSHING

PEOPLE HOLlDAYS BOATS

/~~EATING --- WORK ------ MONEY ----- CARS

Figure 1.15 Exemplary network of cognitive models

not be neglected, namely the fact that cognitive models are ornni-presento

In every act of categorization we are more or less consciouslyreferring to one or several cognitive models that we have stored.Only in the very rare case when we encounter a totally unfarniliarobject or situation will no appropriate cognitive model be avail-able, but even then we will presumably try to call up similarexperiences and irnmediately form a cognitive model. Imagine thecase of a foreign visitor to Britain who goes to watch a cricketgame for the first time in his life. Having no cognitive model ofthe situation at his disposal he will have no idea what is going onon the field. Nevertheless the visitor will try to relate what he seesto similar cognitive models perhaps about games in general orabout his knowledge of baseball.

Clearly, then, we can neither avoid the influence of cognitivemodels nor function without them. Even in the rather artificialsituation of goodness-of-example ratings, the cognitive model ofthe field that the subjects are being asked to rate is at work.Therefore, it would be misleading to say that such experiments takeplace in an uncontextualized vacuum. What these experimentalsituations generate is a very neutral or zero-context which can becompared to sentences like When she looked out cif the window she sawa ... or She thought cif the .... Uncontextualized language in the senseof 'language without cognitive models' is apparently unthinkable.

Cultural models

Cognitive models, as the term suggests, represent a cognitive,basically psychological, view of the stored knowledge about a cer-tain field. Since psychological states are always private and

50 An Introduction to Cognitive Linguistics

individual experiences, descriptions of such cognitive modelsnecessari1y involve a considerable degree of idealization. In otherwords, descriptions of cognitive models are based on the assump-tion that many people have roughly the same basic knowledgeabout things like sandcastles and beaches.

However, as the cricket example has shown, this is only partof the story. Cognitive models are of course not universal, butdepend on the culture in which a person grows up and lives. Theculture provides the background for all the situations that we haveto experience in order to be able to form a cognitive model. ARussian or German may not have formed a cognitive model ofcricket simply beca use it is not part of the culture of his own coun-try to play that game. So, cognitive models for particular domainsultimately depend on so-called cultural models. In reverse, cul-tural models can be seen as cognitive models that are shared bypeople belonging to a social group or subgroup.

Essentially, cognitive models and cultural models are thus justtwo sides of the same coin. While the term 'cognitive model'stress es the psychological nature of these cognitive entities andallows for inter-individual differences, the term 'cultural model'emphasizes the uniting aspect of its being collectively shared bymany people. Although 'cognitive models' are related to cognitivelinguistics and psycholinguistics while 'cultural models' belong tosociolinguistics and anthropologicallinguistics, researchers in all ofthese fields should be, and usually are, aware of both dimensionsof their object of study.

Our earlier reference to Malinowski's 'context of culture' hasalready shown that to include cultural aspects in linguistic con-siderations is not really a recent invention. Yet although thecultural background has long been part and parcel of investigationsin sociolinguistics and anthropologicallinguistics, only few seman-ticists have bothered to deal with such matters.22 In the followingwe will discuss two examples of categories where cultural modelsare highly relevant for a proper understanding of cognitive cate-gories and their structure.

One example of diverging cultural models and the effects ofthese differences on the structure of a category is discussed by Lipka(1987: 292). Consider the two objects depicted in Figure 1.16. Itcan be seen that they differ considerably in their size, height andthe presence or absence of drawers, yet within the right culturalcontext both are prototypical examples of the category DESK. Inorder to understand this, one must know that in China andJapan

Prototypes and categories 51

tTiD- -- -- - fÍ:r-/------rll/(a) (b)

Figure 1.16 Illustration ofEuropean and traditionalJapanese prototypesof the DESK category (taken frorn Lipka 1987: 292)

writing was traditionally performed sitting cross-legged or on one'sheels on the floor. Within this cultural model of writing, object (b)in Figure 1.16 is prototypical while object (a) would be a ratherperipheral member of the DESK category. The opposite is of coursetrue in a modero European contexto In this context one would siton a chair in front of the desk and a prototypical DESK would alsobe used to store stationery, important personal documents and afew other things.

As a second example we will discuss the English and Frenchprototypes of the category FlRST MEAL OF THE DAY. The proto-typical attributes of the meal in the two countries are listed inFigure 1.17. The two lists in Figure 1.17 show how different theFrench and the English prototype of the category are. While theFrench >PETIT DEJEUNER < is a rather frugal affair consisting of alarge bowl of coffee and a croissant, the >ENGLISH BREAKFAST<

includes a whole array of things to eat and drink. Since the Frenchbreakfast needs much less crockery, cutlery and atmosphere, Frenchhotels often do not provide a breakfast room, but serve breakfaston a tray in the bedroom or ask you to have your coffee and crois-sant in a nearby café or bar. In contrast, the English breakfast isnever served in the bedroom, but in a breakfast room. The reasonswhy the two types ofbreakfast are so different, and this is our pointhere, is that they reflect different cultural models concerning thefunction and relevance of the meals of the day. According to theFrench MEALS OF THE DAY model, breakfast is of minor importancebeca use the midday meal is, supposed to be rather solid and will befol1owed by another súbstantial evening meal. In contrast, theEnglish cultural model is based on a substantial first and last meal,while the midday meal or lunch is a rather slender affair.

What this example also illustrates is that cultural models arenot static but changing. In fact, what we have described as theEnglish prototype may still be practised in hotels and bed and

Naive models and expert models

Being based on our everyday experiences of the phenomenaaround us, cultural models may include assumptions that, from astrictly scientific point of view, may be questionable or eveninaccurate. For example, it is part of our cultural model of 'dogs'that wagging their tails means '1 am happy' and growling means '1am angry'. When cats are purring they are said to be enjoyingthernselves. For an ethologist or biologist, all these beliefs are highlydubious, beca use it is by no means clear that animals have emo-tions at all in the sense that humans do. In their expert modelswhich are based on hard scientific facts and the rules oflogic, thesetypes of assumptions would have no place. Naive cultural models,on the other hand, are based on informal observations, traditionalbeliefs, and even superstitions, and have therefore also been called'folk-models' .

For obvious reasons, the discrepancy between the scientific-ally founded models of experts and the naive models oflaypersonsis particularly noticeable in scientific and technical domains.Consider for example the case of the naive model of the physical

52 An Introduction to Cognitive Linguistics

>PETIT DEJEUNER <

Components:coffeecroissant

>ENGLlSH BREAKFAST<

Components:cereal and rnilktea or coffee, orange juicetoast, butter, marmaladebacon, eggs, baked beans, sausages,tomatoes

served at bedsideor local café or bar

served in breakfast room

not included in room rate included in room rate

Figure 1.17 Prototypical attributes of the category FlRST MEAL OF THEDAY in Britain and France

breakfast places, but can no longer be regarded as standard routinein farnilies. Conversely, many continental hotels which used toserve a 'Continental breakfast' modelled on the French prototype(though normally replacing croissants by something else) are nowoffering 'buffet breakfasts', which have many of the attributes ofan English breakfast.

Prototypes and categories 53

phenomenon of motion. McCloskey (1983) carried out exp~ri-rnents and interviews to elicit the cultural model of motionprevalent in America. He asked his informants to imagi.n~ an air-plane flying at constant speed and altitude. In addition, theinformants should assume that at one point during the journey alarge metal ball is dropped frorn the plane, which continues flyingat the sarne speed and altitude and in the same direction. The taskwas to draw the path the ball wi11follow until it hits the ground,ignoring wind and air resistance. Its final position in relation to theplane should also be indicated. Before you read on, you shouldperhaps try to solve the task yourself, i.e. make your own drawingof the paths followed by the plane and the metal ball.

Now compare your drawing with the scientifically correctanswer to the problem. As physicists tell us, the ball wi11fall in akind of parabolic are and hit the ground directly below the pointthe plane has reached in the meantime. The ball will take this kindof path, because it will continue to travel horizontally at the samespeed as the plan e while acquiring constantly increasing verticalvelocity.

You need not be disappointed if your drawing does not agreewith the scientific explanation. In fact, in McCloskey's experimentthe scientifically correct response was only given by 40 per cent ofthe informants. The majority of the subjects thought that the ballwould take a different course, for instance that it would drop in astraight line or would fall in a diagonal. Obviously these 60 percent, plus perhaps yourself, have a naive cultural model of motionthat differs from the expert model current in physics. .

What this experiment shows is that the cultural models heldby the majority of the people need not be, and often are not, inline with the objectively verifiable, scientific knowledge availableto experts. If we consider that cultural models are based on thecollective experience of a society or social group this does notcome as a surprise. To get through everday life, laypersons do notneed scientifically correct models, but functionally effective ones.This means that as long as a model is in line with what we per-ceive and enables us to make functionally correct predictions, itcan have widespread currency although it may be technicallyinaccurate.

Another illuminating example is provided by Kernpton(1987). When she studied the American cultural model of homeheat controls or thermostats by means of interviews and behavi-oural records, she found two cornpeting theories.

54 An Introduction to Cognitive Linguistics

One, the feedback theory, holds that the thermostat senses temperatureand turns the furnace on and off to maintain an even ternperature. Theother, which I call the valve theory, holds that the thermostat controlsthe amount of heat. That is, like a gas burner or a water valve, a highersetting causes a higher rate of flow,

(Kempton 1987: 224)

The feedback theory is technically correct, while the valve theoryis wrong. What is of special interest about the two theories is thateven though the valve theory is wrong, it also enables us to makethe right predictions for the control of temperature in a house andtherefore there is no reason why laypersons should not espouse it.

It seems, then, that many naive cultural models, especially inthe scientific and technological domain, are inaccurate from ascientific point of view, but usually correct as far as their functionalpredictions are concerned. In other domains of everyday life thequestion of the accuracy of a model does not seem to be as rele-vant. For example, for the cultural models OfSANDCASTLE,BEACH,DESKSand BREAKFASTwhich we have singled out in this sectionfor illustrative purposes, it would not be appropriate to speak ofcorrect or inaccurate models, although experts with particularlyrefined cognitive models could certainly be found for all spheres.What counts is that 'ordinary' everyday experiences do not followthe doctrines laid down for scientific research and the rules offorrnallogic, but have other, more genuinely cognitive, principlesbehind them, some of which will be discussed below in Chapters3 and 4.

To conclude this section let us surnmarize the main issues thatwe ha ve addressed here:

(1) Cognitive categories interact with and influence each otherand this can cause a shift of category prototypes, of bound-aries and of the whole category structure.

(2) Over and above the actual context in which the use ofcategories is embedded, the internal structure of categoriesdepends on cognitive and cultural models which are alwayspresent when language is processed.

(3) A number of terrninological distinctions seem necessary for adifferentiated view of the context-dependence of categories.Thus we have defined

situation as the interaction of objects in the real world;context as the cognitive representation of the interactionbetween cognitive categories (or concepts);

(4)

Prototypes and categories 55

cognitive model as the sum of the experienced and sto redcontexts for a certain field by an individual;cultural model as a view of cognitive models highlightingthe fact that they are intersubjectively shared by the mem-bers of a society or social group.

'Naive' cultural models, especially those for technical domains,need not be in line with the scientifically accurate knowledgeof experts, but may be based on what is cornmunal experience,and strictly speaking even 'wrong' assurnptions. Neverthelessthese naive cultural models can be shared by most laypersonsin a society as long as the functional predictions they make arecorrecto

Exercises

1. In pragmatics and sociolinguistics the participants of a speechevent are often seen as part of the wider 'situational context'.Discuss this notion of'context' in relation to the one put for-ward in this chapter.

2. Object categories like CAR are characterized by attributesrelating to their forrn, size, material, parts, functions, and theassociations and emotions they call up. Discuss which of theseattributes are more likely to change their 'weight' when thecontext changes, let us say from ordinary traffic to a car racecontexto

3. Repeat the two-stage test in exercise 5 of section 1.1 withspecial contexts like The estate agent climbed out if his ... (Jaguar,Rolls Royce, BMW, Mercedes, etc.) or The children loved to dimb the... (apple tree, pear tree, cherry tree, etc.) in the orchard given beforethe association and the goodness-of-example ranking task.

4. Eskimos have many words for different types of snow,Aborigines for different types of sand, and in Arabic one mustchoose from a whole range of words which are subsumedunder the Western category CAMEL(cf. Lyons 1981: 67). Canyou explain these phenomena with the help of the notion of'cultural model'?

5. Compare the cultural model BACHELORwith that of its appar-ent counterpart SPINSTER. Discuss the parallel examplesGENTLEMAN-LADY,MASTER-MISTRESSand BOY-GIRL.

6. Ask your friends and farnily how they imagine the variousprocesses carried out by the components of a personal

56 An Introduction to Cognitive Linguistics

computer and extract the naive cultural model from theiranswers. If you happen to know any specialists in the field,compare the naive model to the expert model.

7. We found that naive cultural models are regarded as valid aslong as they make the right predictions. Look for linguisticexamples that reflect either wrong or outdated naive modelslike the sun rises in the east (of course the earth revolves) orthe apple Jell to the ground (it is attracted by the force ofgravitation). Discuss examples of so-called folk-etymologieslike sparrow-grass for asparagus, causeway for chausée, cowcumberfor cucumber, and the compounds crayfish, starJish and jelly-

fish.

Suggestions for further reading

Section 1.1

For recent accounts of the early development of research intocategorization see Lakoff(1987: ch. 2) and Taylor (1995: ch. 3.2).Kleiber (1990) provides a discussion of the development of proto-type theory, though on a fairly abstract level. Schwarz (1992) alsogives a short overview (in German), but on the whole concentrateson what we have called the 'logical view' in cognitive linguistics(see Introduction). The shortest description of Rosch's importantcontributions to the field is available in Rosch (1978). Rosch(1988) is an informal retrospect of the early stages of prototypetheory from her own point of view, which is definitely worthreading.

1. Ullmann (1962: 116-28) discusses the sources of vaguenesswith many examples from different languages. Quine (1960:125f) puts forward some interesting thoughts on vaguenessfrom a philosophical perspective. More recent publications onthe topic from a cognitive perspective are Geeraerts (1993)and Tuggy (1993).

2. Lakoff (1987: ch. 18) discussesWhorf and relativism in somedetail. Wider philosophical implications of relativist theoriescan be found in Putnam (1981: 103-26, esp. 119-24).

3. The introductory section ofBerlin and Kay (1969: 1-5, 14fT)certainly makes good reading. A concise survey of variousreactions to Berlin and Kay (1969) is provided by Taylor(1995: 10ff). Kay and McDaniel (1978) present neurophysio-

Prototypes and categories 57

logical data which support part ofBerlin and Kay's claims.Fora recent re-interpretation of Berlin and Kay's findings seeWierzbicka (1990).

4. Smith and Medin (1981) provide a survey of various theoriesof categorization. Though not the most recent psycholin-guistic textbook, Clark and Clark (1977) is still helpful for thebasic psychological and psycholinguistic aspects of categoriza-tion. Aitchison (1994) focuses on the storage of categories inthe 'mentallexicon'.

5. Up to 1972, E. Rosch published under her earlier narneHeider. So Heider (1971, 1972) and Heider and Oliver (1972)are written by the same author as the publication in thebibliography under the name of Rosch.

6. Lyons (1977: 109ff) is helpful to get a first idea of realism,nominalism and conceptualism, i.e. theories of status of theworld. Discussionsof the realist-conceptualist model that pro-vides the basisfor much thinking in the paradigm of cognitivelinguisticscan be found in Johnson (1987: ch. 8) and in Lakoff(1987: ch. 16). Both chapters are partially based on Putnam(1981: esp. ch. 3) which makes some fascinatingbut not quiteeasy reading.

7. Being the later article, Labov (1978) is more comprehensive;however, the developments mainly concern the mathe-matical technicalities involved in formalizing fuzziness.

8. For discussions of polysemy and the problem of keeping itapart from homonymy in the structuralist tradition see Lyons(1977: 550ff), Leech (1981: 227ff), Lipka (1992: 135ff). Forthe lexicographical treatment of polyserny from a prototypeperspective see Geeraerts (1990).

Section 1.29. For a psychologically oriented account of the classicalview

of categories see Smith and Medin (1981: ch. 3); for a des-cription from a linguistic and philosophical point of view,Lakoff (1987: ch. 11). Lakoff (1987) also has a chapter (ch. 9)on 'Defenders of the classical view', which in fact hasthe function of defending his own prototypical view ofcategories. As regards the notion of 'essential features', theclassicis Katz and Postal (1964: ch. 2). Lipka (1992: ch. 3.3)gives a concise introduction which also integrates non-necessaryfeatures.Wierzbicka (1985) considerably extends the

58 An Introduction to Cognitive Linguistics

notion of essential features and gives interesting, but verydetailed, feature definitions for organisms and everydayobjects.

10. For details on the statistical method used see Rosch andMervis (1975) and Rosch et al. (1976). These articles are alsointeresting because the statistical notion of 'cue validity' isintroduced to measure what we call attribute-based ratings.Explanatory remarks are provided in Smith and Medin (1981:78ff) and Geeraerts (1988a).

11. For a similar view of categorization illustrated with the cate-gory UNCLE see Langacker (1987a: 19ff).

12 The most convenient access to gestalt psychology is via not-too-recent histories of psychology, e.g. Boring (1942) andBoring (1950: ch. 23); the second book also provides inter-esting information on the development of the theory andthe biographies of its major proponents Wertheimer, K6Werand Koffk:a. For original sources of 'gestalt', 'Pragnanz' and'gestalt principles' see Koffk:a (1935) and Wertheimer(1958).

13. The function of parts of objects in categorization is discussedin a number of publications by Tversky and Hemenway (seeTversky and Hernenway 1984; Tversky 1990).

14. Illustrations in dictionaries are discussed by Stein (1991) andLipka (1995), though from a lexicographic rather than a cog-nitive point of view. For an account of psychological researchinto the perception of shapes using 'templates' see Smith andMedin (1981: ch. 6).

15. Aitchison (1994) is a very palatable survey of empirical andtheoretical approaches to the mental lexicon, which is pre-dominantly based on linguistic and psycholinguistic work. Acritical discussion from a psychological point of view can befound in Smith (1978).

16. Opinions about how to interprete typicality effects havechanged with the development of the prototype model. ForRosch's later position see Rosch (1978) and (1988). For moreextensive discussionsof the issue, in particular for the distinc-tion between typicality and membership (which we findunnecessary) see Lakoff (1987: 40ff and ch. 9) and Kleiber(1990). Geeraerts argues that prototypicality itself should beregarded as a prototypical notion (1988a) and discussespoten-tial sources of prototypicality (i988b).

Prototypes and categories 59

Section 1.317. For an empirical examination of the context-dependence of

prototypes and the category structure see Roth and Shoben(1983), who investigated the effects of sentences like The birdwalked across the barnyard on the internal structure of thecategory BIRD.

18. The closest approximation to our conception of the terms'context' and 'situation' that we know of is van Dijk (1981:222, 269 et passim) who emphasizes the selective or 'abstrae-tive' nature of the cognitive contexto

19. An attempt to provide experimental support for a distinctionbetween context-dependent and context-independent infor-mation in categories can be found in Barsalou (1982).

20. For notions which largely correspond to our conception of'cognitive models' see Johnson (1987: esp. 28f, 10lff) on'image-schemata'; Lakoff (1987: 68ff, 118ff) on 'idealizedcognitive models'; Gentner and Stevens (1983) and Johnson-Laird (1983) on 'mental models'; and Holland and Quinn(1987) on 'cultural' and 'folk models'. Langacker's notion of'domains' is discussed in Chapter 4, section 4.3, and thenotion of 'frames' in Chapter 5, section 5.1.

21. A good introduction to the connectionist theory of the mindis provided by Bechtel and Abrahamsen (1991).

22. For German-speaking readers interested in differencesbetween cultural models current in Britain and on theContinent, a very interesting and readable source 'is Leisi(1985: 87ff). He discussesthe British models for such spheresas the ACADEMIC FIELD, HOUSEHOLD ITEMS, FOOD and SPORTS.

AlI rights reserved; no part of this publication may be reproduced, stored in a retrievalsystem, or transmitted in any form or by any means, electronic, mechanical,photocopying, recording, or otherwise without either the prior written pennission of thePublishers or a licence pennitting restricted copying in the United Kingdom issued bythe Copyright Licensing Agency Ltd., 90 Tottenham Court Road, London WIP 9HE.

Addison Wesley Longrnan Lirnited,Edinburgh Gate,Harlow, Essex CM20 2JE, Englandolld Assoaated COlllpollies throughout the world.

Published in the United Sto/es of Americaby Addison Wesley LoIIgmoll, New York

© Addison Wesley Longman Limited 1996

First published 1996

ISBN 0--582-239664 PPR

British Library Cataloguing-in-Publication Data .

A catalogue record for chis book is available from the British Library

Library of Congress Cataloging-in-Publication Data

Ungerer, Friedrich.An introduction to cognitive linguistics / Friedrich Ungerer,

Hans-Jiirg Schmid.p. cm. - (Learning about Ianguage)

Includes bibliographical references and index.ISBN 0--582-23966--41. Cognitive gramrnar. 1. Schrnid, Hans-Jiirg. 11. Title.

11I. Series.P165.U54 1996415--dc20 95-48497

CIP

Set by 8 in Bembo 11/12pt

Produced through Longman Malaysia, TCP

Contents

Typographicalconventions VII

Acknowledgements VIII

Introduction X

1 Prototypes and categories 11.1 Colours, squares, birds and cups: early empirical

research into lexical categories 11.2 The internal structure of categories: prototypes,

attributes, family resemblances and gestalt 201.3 Context-dependence and cultural models 43

2 Levels of categorization 602.1 Basic level categories of organisrns and concrete

objects 602.2 Superordinate categories and experiential

hierarchies 732.3 Subordinate categories, composite terms and

word-formation 862.4 Basic level categories and basic experiences: actions,

events, properties, states and locations 99

3 Conceptual metaphors and metonymies 1143.1 Metaphors and metonyrnies: from figures of speech to

conceptual systems 1143.2 Metaphors, metonyrnies and the structure of emotion

categories 1303.3 Metaphors as a way of thinking: examples from

science and politics 143

vi Contents

4 Figure and ground 1564.1 Figure and ground, trajector and landmark: early

research into prepositions 1564.2 Figure, ground and two metaphors: a cognitive

explanation of simple clause patterns 1714.3 Other types of prominence and cognitive

processing 187

5 The frame and attention approach 2055.1 Frames and scripts 2055.2 Event-frames and the windowing of attention 2185.3 Language-specific framing and its use in

narratives 233

6 Other issues in cognitive linguistics 2506.1 Iconicity 2506.2 Grarnmaticalization 2556.3 Lexical change and prototypicality 2606.4 Effects on foreign language teaching 267

Conclusion 278

References 281

Index of persons 294

Index of subjects 297

Typographical conventions

Cognitive categories small capitalse.g. BIRD, VEHICLE

Attributes single quotese.g. 'juicy', 'has legs'

Members of categories arrows and small capse.g. >ROBIN<, >PARROT<

Cognitive/ cultural models underlined small capse.g. ON THE BEACH, ¡OURNEY

Metaphors/ metonyrnies + signs and small capse.g. +ANGER IS HEAT+

Image schemas - signs and small capse.g. -BE IN-, -PATH-

small italicised capse.g. DOC, CHAIR

small caps in bracketse.g. [COMMERCIAL EVENT]

Domains

Frames