From Polyseme to Ling ChangeFrancoisSemanticMaps

7/30/2019 From Polyseme to Ling ChangeFrancoisSemanticMaps

1/29

2nd proofs

Volume 106

Fro Polyy to Stc Chg. owrd tyolog y

o lxcl tc octo

Edtd y Mrt Vhov

Editors

Werner AbrahamUniversity o Vienna

Michael NoonanUniversity o Wisconsin, Milwaukee

Editorial BoardJoan BybeeUniversity o New Mexico

Ulrike ClaudiUniversity o Cologne

Bernard ComrieMax Planck Institute, LeipzigUniversity o Caliornia, Santa Barbara

William CrofUniversity o New Mexico

sten DahlUniversity o Stockholm

Gerrit J. DimmendaalUniversity o Cologne

Ekkehard KnigFree University o Berlin

Christian LehmannUniversity o Erurt

Robert E. LongacreUniversity o Texas, Arlington

Brian MacWhinneyCarnegie-Mellon University

Marianne MithunUniversity o Caliornia, Santa Barbara

Edith MoravcsikUniversity o Wisconsin, Milwaukee

Masayoshi ShibataniRice University and Kobe University

Russell S. TomlinUniversity o Oregon

Studies in Language Companion Series (SLCS)

T r h tlhd coo r to th rodcl

Studies in Language.

From Polysemy

to Semantic Change

owrd tyology o lxcl tc octo

Edited by

Mrt VhovLlc (Ilco, CNRS), Fdrto UL

Joh Bj Plhg Coy

Atrd / Phldlh


2/29

2nd proofs

Library of Congress Cataloging-in-Publication Data

Fro olyy to tc chg : towrd tyology o lxcl tc octo

/ dtd y Mrt Vhov.

. c. (Std Lgg Coo Sr , issn 0165-77 63 ; v. 106)

Icld logrhcl rrc d dx.

1. St c, Htorcl. 2. Polyy. 3. yolo gy (Lgtc ) I . Vhov,

Mrt.

P325.5.H57F76 2008

401'.43--dc22 2008031821

isbn 978 90 272 0573 5 (H; lk. r)

2008 Joh Bj B.V.

No rt o th ook y rrodcd y or, y rt, hotort, crofl, or y

othr , wthot wrtt ro ro th lhr.

Joh Bj Plhg C o. P.O. Box 36224 1020 me Atrd T Nthrld

Joh Bj North Arc P.O. Box 27519 Phldlh pa 19118-0519 usa

T r d th lcto t th rqrt o

Arc Ntol Stdrd or Iorto Scc Prc o

Pr or Prtd Lrry Mtrl, ansi z.-.

8TM

able of contents

Semantic associations: A oreword vii

Martine Vanhove

Part 1. State of the art

Approaching lexical typology 3

Maria Koptjevskaja-Tamm

Part 2. Teoretical and methodological issues

Words and their meanings: Principles o variation and stabilization 55

Stphane Robert

Te typology o semantic anities 93

Bernard Pottier

Cognitive onomasiology and lexical change: Around the eye 107

Peter Koch

Mapping semantic spaces: A constructionist account o the light verb

xordaen eat in Persian 139

Neiloufar Family

Semantic maps and the typology o colexication: Intertwining polysemous

networks across languages 163

Alexandre Franois

A catalogue o semantic shis: owards a typology o semantic derivation 217

Anna A. Zalizniak

Semantic associations and confuences in paradigmatic networks 233

Bruno Gaume, Karine Duvignau & Martine Vanhove

Part 3. Case studies

About Eating in a ew Niger-Congo languages 267

Emilio Bonvini


3/29

2nd proofs

Semantic maps and the typology

o colexifcation

Intertwining polysemous networks

across languages

Alexandre FranoisLacito (CNRS), Fdration UL

Building upon the model o Semantic Maps (Haspelmath 2003), which

typologists have designed mainly or grammatical semantics, this chapter

discusses methodological issues or a model in lexical typology.

By breaking up polysemous lexemes o various languages into their semantic

atoms or senses, one denes an etic grid against which cross-linguistic

comparison can be undertaken. Languages dier as to which senses they

colexiy, i.e., lexiy identically. But while each polysemous lexeme as a whole

is language-specic, individual pairings o colexied senses can be comparedacross languages. Our model, understood as an empirical, atomistic approach to

lexical typology, is nally exemplied with the rich polysemies associated with

the notion breathe. Intertwined together, they compose a single, universal

network o potential semantic extensions.

Keywords: breathe; colexication; etic grid; lexical typology; methodology;

polysemy; semantic maps; sense; soul; spirit

1. General issues o lexical typology*

At rst sight, the capacity o the human brain to detect analogies in ones environment

is innite, and should logically result in lexical polysemy having no limits. And indeed,

the more languages we explore, the more examples we nd o unique metaphors and

unexpected cases o semantic shi probably one o the most thrilling mysteries

and charms o language discovery. But what generally happens is that we ocus our

*I would like to thank Martine Vanhove, Maria Kotjevskaja-amm, Sergue Sakhno and Fran-oise Rose or their precious comments on a previous version o this article.


4/29

2nd proofs

1 Alexandre Franois

attention on the most exotic cases, and overlook the inormation that is o most inter-

est or the hunter o semantic universals: namely, that a great deal o lexical polysemies

are in act widespread across the worlds languages, and, as such, deserve to be high-

lighted and analyzed.

Tis observation meets the agenda o lexical typology. Indeed, one o the main-

stays o typological linguistics is precisely to show that cross-linguistic variation, ar

rom being random and innite, can in act be reduced to a limited range o possible

cases. And it is the purpose o this whole volume to show that the search or univer-

sals and typological tendencies, which has already proven ruitul in phonological or

grammatical studies, may perectly apply to the study o the lexicon too, provided the

specic methodological issues it raises are properly addressed.

Generally speaking, one central issue raised by linguistic typology is the neces-

sity to ascertain the comparability o languages. Languages can only be contrasted

with accuracy provided a standard o comparison is proposed, dening the common

ground against which commonalities and dierences can be observed across lan-

guages. Studies in grammatical typology have already begun to identiy some o the

relevant criteria or the comparison o grammar systems. Tey consist in the many

unctional eatures that emerge out o the observation o actual categories in natural

languages: such notions as number, animacy, deixis, telicity, agentivity , orm a solid

unctional basis or the cross-linguistic analysis o specic points o grammar. But inthe less explored domain o lexical typology, the comparability o languages seems less

easy to delineate.

Several reasons may account or this scientic gap. For one t hing, there is still the

widespread idea that grammars are tidy and regular, while lexicons would be open-

ended, exuberant and idiosyncratic. With such a perception, it is deemed unlikely that

the typological project might come up with any satisying generalizations in the lexical

domain as much as it does in the observation o grammars. Also, the accurate descrip-

tion o lexical data oen requires taking into account the many unctional properties

o real-world reerents, to say nothing o the pitalls o culture-specic vocabulary; this

seems to make the comparative project a dicult challenge.

Te aim o this article is to discuss and illustrate the possibility o comparing the

worlds lexicons, by resorting to a methodological tool which has already proven its

eciency among grammar typologists: semantic maps. For a given notion taken as

the maps pivot, I will suggest a method or drawing a universal network o poten-

tial semantic extensions, ollowing the observation o polysemies attested across the

worlds languages. A useul concept or this study is the notion ocolexication, which

will be introduced in 3.2. Finally, the last part o this paper will illustrate the poten-

tials o this method, by analyzing the complex semantic network associated with the

notion breathe.

Semantic maps and the typology o colexication 1

2. Ensuring the comparability o lexicons

2.1 Monosemy vs. polysemyTe rst issue that has to be addressed when studying the lexicon, is the nature o

the objects to be compared. Indeed, the comparative project will be directly aected

by theoretical choices regarding the nature o the word, whether it is understood as

intrinsically monosemous, or capable o genuine polysemy. Tis discussion relates toongoing debates (see Geeraerts 1993; Nerlich et al . 2003; Riemer 2005) which I will

only mention briefy here in relation to the present discussion.

When Saussure dened the sign as the arbitrary pairing o a orm (the signier)

and a concept (the signied), he insisted that each concept can only be characterized

negatively, insoar as it contrasts with other words o the same language: Concepts

() are purely dierential; they are dened not positively by their contents, but nega-

tively by their relationship to the other elements o the system.1 Tis conception o

semantics has led to the structuralist view that the meaning o a given word in one

language will never match exactly the meaning o its most usual translation in another

language: its semantic outline, as it were, is unique to that particular system, and can-

not be ound identical anywhere else. In such a ramework, the very project o a lexical

typology, aiming to compare lexicons across languages, seems not only dicult, but

simply out o the question.

Directly inherited rom this structuralist st andpoint is the monosemist approach,

whereby a polysemous lexical unit will be analyzed as undamentally organized

around a unique general meaning; its dierent attested senses in context are under-

stood as resulting rom the combination o that core meaning with the pragmatics

o each specic speech situation. Conversely, the polysemist approach considers t he

multiplicity o meanings to orm an intrinsic property o each polysemous word at the

semantic level, with no necessity, or even legitimacy, to reduce this multiplicity to an

articial unity. Several attempts have been proposed to reconcile these two contrary

approaches, or example, around the notions o prototype (Rosch 1973) or radial

categories (Lako 1987).

It is not the purpose o this article to solve such long-discussed issues. What is rel-

evant here is to underline that each point o view is an attempt to handle the dialecticbetween unity and multiplicity, which is inherent to the paradox o polysemy. Now, it

appears that cross-linguistic comparison can be carried out with more precision i t he

1. Les concepts () sont purement direntiels, dnis non pas positivement par leur

contenu, mais ngativement par leurs rapports avec les autres termes du systme (Saussure

1972 [1916]: 162).


5/29

2nd proofs

1 Alexandre Franois

acts o polysemy are stated explicitly rom the perspective o a multiplicity o senses.

Te idea that each polysemy is undamentally underlied by a single abstract mean-

ing, though intellectually appealing it may be, results in denitions that are dicult

to apprehend with precision, and to test against actual data. o quote the words o

Haspelmath (2003: 214), general-meaning analyses are not particularly helpul i one

wants to know in what way languages dier rom each other.

Whatever theoretical viewpoint one adopts concerning polysemy, the only rep-

resentation that really allows cross-linguistic comparison is thereore one that explic-

itly spells out the multiplicity o senses making up a words polysemy. Te question

whether these senses are to be understood as pragmatically dened contextual uses

o a central meaning (monosemist approach), or as autonomous components at the

semantic level (polysemist approach), is somewhat a secondary issue. What is essential

is to nd a method t hat will allow us to describe each polysemous network in the ull

detail o its internal components.

2.2 Overlapping polysemiesA rst illustration can be proposed, with the English word straight. Roughly speaking,

this adjective may be broken into at least the ollowing senses2 (see 3.1. or a discussion

o the method):

rectilinear (a straight line) heterosexual (gay or straight)

rank (straight talking) undiluted (straight whisky)

honest (a straight guy) directly (straight to the point)

classical (a straight play) immediately (straight away)

Its closest translation in French, droit, shows a slightly dierent range o senses:

rectilinear (un trait droit)

directly (aller droit au but)

honest (un type droit)

right-hand (le ct droit)

Now, a strictly monosemist approach would probably try to dene the core meaning

ostraightby resorting to a general denition, suciently abstract so as to encompassall its contextual uses in English. Ten it would also propose a unique denition or

French droit; and because the meanings attested or these two words are so close to

each other, it is likely that the t wo general denitions would end up being quite similar,

2. Troughout this paper, angled brackets are used to represent senses, insoar as they

orm an element o a polysemous network.


and thereore unable to grasp clearly what is common and what is dierent between

straightand droit. Te comparison becomes much easier and clearer i the comparison

is carried out at the level o the senses. It is then easy to observe that the two words

share exactly three senses: rectilinear, directly, honest; that French droitadds to

these a sense right-hand, while English straightadds a number o other senses which

have no equivalent in French.

Tis conguration may be illustrated visually in the orm o two overlapping

sets (Figure 1). Te elements o the sets are the senses, presented here in no spe-

cic order. Te sets themselves reer to the lexical units the words that happen

to group these senses in their own polysemies. One may talk here o two overlap-

ping polysemies.

right-handrectilinearhonestdirectly

undilutedclassicalheterosexualimmediately

frank

Fr. droitEng. straight

Figure 1. Overlapping polysemies: Eng. straightvs. Fr. droit.

In sum, the ne-grained comparison o lexicons across the worlds languages can

be ecient provided each polysemous network is rst broken down into its seman-

tic atoms or senses. Tis may be done regardless o ones theoretical preerences

whether these senses are taken as actual s emantic sub-categories in the speakers minds,

or merely contextual maniestations o a deeper meaning. Tis approach, whereby a

given word is analyzed into its semantic atoms, is the rst step beore languages can

be compared with precision, showing which senses each language lexies together. In

this new perspective, the primary unit o observation or lexical typology is no longer

the word a complex, highly language-specic entity but the sense a unctionally-

based, language-independent criterion (3.1.).

Tese observations orm the basic principles o the model I will introduce in the

remainder o this article. Section 3 will rst discuss the methodology or isolating senses,

and or observing the way languages group them together; I will then introduce the con-

cept o colexication. Section 4 will discuss the principles underlying the representation

o lexical semantic maps, drawing on the principles set out by Haspelmath (2003). Te

model here delineated should provide empirical tools or the observation and analysis o

polysemy across languages. Hopeully, it should also make it possible or uture research


6/29

2nd proofs

1 Alexandre Franois

to detect certain typological tendencies among the lexical structures o the worlds

languages, and eventually pave the way or the ormulation o lexical universals.

3. Towards a typology o colexifcation

For each specic notion taken as the object o study (see 4.3.), the empirical method

here adopted ollows two steps:

First, select the word that lexies this notion in one language, and identiy the

various senses which orm part o its polysemy, in this particular language.

Second, once a list o senses has been proposed or this rst language, observe a

second language, to see which o these senses are also lexied together ( or colexi-

ed), and what new senses have to be added to the list. Ten proceed to another

language, and expand the list accordingly.

o use a chemical metaphor, one could say that the comparison o dierent molecules

requires rst to identiy the nature o the atoms that t ake part in their structure (3.1.);

and then, once each molecule has been broken up into its components, to observe the

bonds that connect these atoms together (3.2.).

3.1 Senses: Te atoms within each moleculeImagine we want to observe the various polysemies attested cross-linguistically around

the notion rectilinear. Te rst step is to select, in any language, a word that may trans-

late (lexiy) this notion; or example, English straight. What now has to be done, beore

being able to compare it with a word rom another language or with another word o

the same language is to break down this lexical unit into its own various senses.

Most o the time, t his is done intuitively, as probably most dictionaries do: obvi-

ous unctional considerations seem sucient to analyze, say, rectilinear and rank

as two distinct senses, deserving separate treatment. However, on some occasions one

may object to the arbitrariness o such intuitive choices, when t wo senses appear to be

so close, that their distinction might be an artiact o t he linguists analysis. In the case

ostraight(2.2.), or instance, one may argue that the psychological senses rank and

honest orm in act a single meaning or the native speaker o English, so that we are

dealing with a case o vagueness rank, honest rather than a case o polysemy, strictly

speaking, between two separate senses.

rying to resolve such a tricky debate with a denite answer might result in unver-

iable and irreconcilable points o view. Luckily, there is one way out o this dilemma,

which is to base all sense distinctions upon the empirical observation o contrasts

between languages. For example, the act that French lexies rectilinear with honest

but not with rank suces to justiy the choice o distinguishing between the two


latter meanings as i they were two separate senses. Even though this may ail to rep-

resent aithully the language-internal perception o an English native speaker, at least

this serves eciently the purpose o cross-linguistic comparison: it becomes then e asy

to state the acts by saying that these two senses are treated the same in English, and

not in French. Te same reasoning would apply to directly and immediately, which

despite their semantic closeness, must be distinguished due to the dierent treatment

they receive in French. Te repetition o the same procedure, or each word under

scrutiny, makes it possible to dene with precision the list o its possible senses.

Tis empirical method o dening senses based on cross-linguistic comparison

has the valuable advantage that it helps sidestep the vexing problem o distinguishing

between polysemy and vagueness (Haspelmath 2003: 231). Now, a corollary o this

approach is that the list o senses or a given word is likely to evolve during the process

o cross-linguistic comparison. Indeed, the more languages are considered, the more

new distinctions are likely to be ound, thereby resulting in the need to split up certain

senses that were initially not distinguished. For example, suppose the examination o

nine languages showed the meaning horizontally rectilinear to be always lexied in

the same way as vertically rectilinear: this would result in the initial grouping o these

two meanings as a unique vague sense rectilinear (horiz. or vertic.), with no empiri-

cal reason or splitting it in two. But once a tenth language is considered that orces to

make this distinction, then the ormer sense rectilinear will have to be cracked downinto two separate senses, or the purpose o cross-linguistic comparison. As a result,

the description given or each polysemous lexeme in the rst nine languages may have

to be revised, due to the introduction o a new semantic distinction aer the tenth

language has been examined.

Note that this remark is not necessarily an issue or the semantic analysis itsel:

one will simply have to describe horizontally rectilinear and vertically rectilinear

as two potentially separate senses, which simply happen to be ormally indistinct in

the rst nine languages, but distinguished in the tenth. Te problem rather arises

at the practical level, i one t hinks o setting up a typological database: or it means that

the semantic descriptions made at a given point in time, during the constitution o the

database, are likely to evolve as more and more distinctions are considered rom new

languages. Tis can entail the necessity or the rst languages entered in the database

to be reassessed again and again as the list o descriptive senses grows. When this takes

the orm o a semantic map (section 4), this also means our maps will have to integrate

the capacity to evolve constantly, and adapt to whatever new input comes in. Tis is

probably easible, but likely to raise certain technical questions.3

3. In the grammatical domain, Haspelmath (2003: 231) reassures us on this point, by saying:

the typical experience is that aer a dozen languages have been examined, ewer and ewer


7/29

2nd proofs

1 Alexandre Franois

Despite these potential issues on the practical side, it is important to see that this

method, by basing every semantic distinction on empirical data, provides a sae anti-

dote against the vagaries o intuition; it ensures that the whole process o semantic

analysis is always veriable and t hereore alsiable.

3.2 Colexication: Te bonds between the atomsIn itsel, the result o the preceding step pretends to be little more than a list o notions

(senses). For one thing, these notions can be shown using t he cross-linguistic method

described above to be unctionally distinct rom each other; but at the same time,

the way they were compiled implies that they are potentially linked together in at least

some o the worlds lexicons. No particular claim is being made at this stage, except

that this non-arbitrary selection o notions should provide a useul etic grid against

which language-specic, emic categorizations are to be observed.

But what is really relevant to our typological study is not so much these atoms

per se, as the bonds that each particular language creates between them. Once a list o

senses is arrived at, the phenomenon most relevant or the second stage o obser vation

may be called colexification .

(1) Agivenlanguageissaidtocolexifytwounctionallydistinct sensesi,and

only i,itcanassociate themwiththesamelexicalorm4

For example, Figure 1 showed that English colexies the senses immediately and

undiluted; rectilinear and right-hand are colexied in French; rectilinear and

directly are colexied both in English and in French. One o the advantages o the

term colexication, which I am proposing here, is to be purely descriptive, and neutral

with respect to semantic or historical interpretations contrary to the term semantic

shi, chosen or example by Anna Zalizniak (this volume).

One interest o the colexication model is to be readily exploitable or typological

research. For example, one may want to check what proportion o the worlds lan-

guages colexiy the two senses rectilinear and honest, as French and English do: is

this connection ound only in a ew scattered languages? Is it an areal phenomenon

covering, say, Western Europe? Is it well represented in other parts o the world? Or isit universally common?

unctions need to be added to the map with each new language. It remains to be seen whether

this comorting statement also applies to the richer realm o lexicons.

. Te term lexical orm may reer to a lexeme or a construction, or occasionally to a lexical

root (but see below or a discussion).


Incidentally, because the list o s enses is initially based on the polysemy o a spe-

cic word in a given language, it is logical that the rst stage o the observation will

show these senses to be colexied in the language under consideration. For example,

because the initial sense list was built as the description o English straight, then it

necessarily results that these senses are all colexied in English. At this stage o the

research, due to a bias in avour o the language taken as the starting point, such an

observation is circular, and has little interest. But these representations become rapidly

more inormative as other languages are considered. For instance, French adds to the

list a new sense right-hand, which is not lexied by English straight. As more lan-

guages are explored, and the list aggregates more and more senses, it will eventually

come closer to a universal grid o potentially interconnected notions with less and

less risk o an ethnocentric bias in avour o a specic language.

3.3 Strict vs. loose colexicationStrictly speaking, the notion o colexication should be understood as the capac-

ity, or two senses, to be lexied by the same lexeme in synchrony. However,

nothing prevents the model rom being extended, so as to make provision or several

hierarchized? levels o colexication. Tese may include the linking o two senses by

a single lexeme across dierent periods o its semantic history (e.g., droitalso meantright, true in Old French); their association in the orm o doublets (e.g., Fr. droitand

direct), or other etymologically related orms (Eng. straightand stretch); the impact o

lexical derivation (Eng. straightstraighten ; Fr. droit droiture honesty) or com-

position (Eng. straight straightorward); and so on.

Ideally, or the sake o accuracy and uture reerence, the dierent types o ormal

relations should be kept distinct in the representation o the data, e.g., with the use

o dierent symbols. In particular, strict colexication (same lexeme in synchrony)

should be careully distinguished rom loose colexication (covering all other

cases mentioned here). Tis will be done here ormally, in tables (sections 5.2. and 7.2),

with the use o respectively + vs [+] signs; and in maps (section 7.3.), with the use

o solid vs. dotted lines. o take an example, one can represent the colexication o

rectilinear and honest in English as strict colexication, because both can be lexi-

ed with exactly the same orm in synchrony (straight). As or the sense simple, easy

to understand, it can also be said to be somehow part o the lexical eld ostraight, but

only indirectly, through the compound orm straightorward; in other words, English

shows strict colexication between rectilinear and honest, but loose colexica-

tion between rectilinear and simple.

Finally, in the ramework o a typological survey carried out around a specic

notion, I propose that the senses to be included in the universal list and in t he map

derived rom it should ll one condition: that is, they should only include those


8/29

2nd proofs

12 Alexandre Franois

senses that are attested to be in strict colexication in at least one language o the world.

For example, supposing one language L1 were ound where exactly the same orm in

synchrony might translate both senses rectilinear and simple, then this would be

a sucient condition or the latter sense to be included in the sense list associated

with rectilinear. Tis being done, it will be possible to state that some languages

like L1 colexiy these two senses directly (strictly), while others like English

colexiy them only indirectly (loosely), and others again do not colexiy them at

all. Conversely, i no language can be ound where the two senses are strictly colexi-

ed, then it is probably a sae principle to exclude them rom the sense list, to avoid

the risk o widening and blurring indenitely the boundaries o a polysemous

network. Tis principle will be useul, or example, in 5.3.2., when discussing the rela-

tionship between breathe (Latin spro) and die (Latin ex-spro). Indeed, because

these two senses oen show some specic semantic relationship through lexical der-

ivation, it would be tempting to include them in the same sense list, and consider

them as indirectly colexied. However, because no language can be ound or obvi-

ous reasons where these two senses are expressed by exactly the same orm (strict

colexication), it is preerable that the sense die be kept away rom the sense list

o breathe.

3. Interpreting colexicationIn principle, the colexication model itsel consists rst and oremost in stating the

acts that is, detecting and documenting the cases o colexication that are empiri-

cally attested across languages. Te interpretation o these semantic connections,

whether it takes a historical or a cognitive perspective or otherwise, arguably belongs

to another phase o the study.

For each pair o senses s1 and s2, several congurations may come out o the data,

suggesting possible questions or the typological study o the lexicon.

In case the colexication o s1 and s2 appears to be attested nowhere, this

may be because the two senses are directly opposite e.g., rectilinear vs. curved;

cognitively divergent rectilinear vs. slow; or simply unlikely to be related

rectilinear vs. green.

I two senses s1 and s2 are colexied in at least one language, this is normally

the sign setting aside the case o accidental homophony that the human brain

has proven able to perceive these senses as somehow semantically connected. Tis

connection may be direct or indirect, via historical paths that may or may not be still

perceived in synchrony. It is then t he purpose o semantic or etymological studies, to

propose a convincing explanation or that connection: is the relationship between s 1

and s2 a case o metaphor, metonymy, hyperonymy, analogical extension ? Is it pos-

sible to reconstruct the direction taken historically by this extension (rom s1 to s2,


or the reverse)? Is it useul to reconstruct a missing link5 between two senses whose

relation is intuitively opaque?

Sometimes one may want to take the reverse perspective, and try to answer the

question why a language does notcolexiy two senses s1 and s2, that is, treats them

separately, when other languages treat them alike. Most oen, this state o aairs

will be simply considered, just like many other linguistic eatures, to result rom a

chance distribution between languages. In some cases, however, hypotheses may be

proposed that would draw a correlation between a specic case o colexication (or o

non-colexication), and, say, the languages environment. For example, Brown (2005a)

suggests that the colexication ohand arm may be infuenced by the geographi-

cal situation o t he community. According to him, the use o tailored clothing cover-

ing the arm in colder environments tends to make the contrast between the hand

and the arm more salient, thus avoring the existence o two separate lexical items.

Likewise, Brown (2005b) sees another correlation between the lexical distinction

nger hand and cultural practises in terms o armers vs. hunter-gatherers.6

Regardless o the likelihood o these hypotheses, it is instructive to see that the acts o

colexication may receive various sorts o unctional explanations, whether semantic,

historical, cognitive or cultural thereby opening ascinating debates.

It may be a subject or discussion, how one should interpret the statistics o colexi-

cation. Tat is, supposing the colexication o s1 and s 2 is particularly widespread inthe worlds languages, should we see this as a sign that these two senses are particularly

close? that t heir semantic connection is unctionally or cognitively particularly

tight? Tis brings in the intuitively appealing notion o degrees o closeness in the

semantic connection. For example, supposing rectilinear and honest turned out to

be statistically more oen colexied than, say, rectilinear and right-hand, one may

think that the rst pair o senses is more deeply motivated than the second pair (?).

Admittedly, however, it may be debatable w hether semantic closeness should be mea-

sured, as I am tentatively proposing here, on statistics based on actual colexication

data rather than assessed, say, on the basis o each notions ontological properties.

. For example, the spatial notion rectilinear is metaphorically associated with social nor-

mality (c. the straight and narrow), as opposed to eccentricity or originality; hence such sensesas classical, not homosexual, not on drugs, etc. In this case, the meaning satisying the

social norm could be described as the missing semantic link whether in diachrony or in

synchrony between several members o this polysemous network.

. Languages o armers tend more strongly to lexically distinguish nger rom hand than

those o hunter-gatherers, which tend more strongly to use a single term to denote both nger

and hand (Brown 2005b: 527). Browns rather unconvincing hypothesis resorts to the saliency

o the nger in those societies which make use o nger rings; he claims that this cultural habit

is more developed among armers.


9/29

2nd proofs

1 Alexandre Franois

Crucially, the descriptive concept o colexication lends itsel to just the same sort

o observations, tests and representations as any other language eature. For example,

specic pairings o senses may be represented in the orm o geographical maps, using

isoglosses or coloured spots.7 Instances o colexication may be attributed to genetic

subgroups and protolanguages (see Franois orthcoming), or result rom local inno-

vations. Tey may also be borrowed through language contact, and take part in areal

phenomena. Tey may change through time, be subject to analogical levelling, and so

on and so orth.

o take just one example, the senses hear and eel are colexied in several areas

o the world: Catalan sentir, Italian sentire, Mwotlapyonteg, Bislama harem:

Knowing that Latin lexied distinctivelysentreeel and audre hear, Catalan and

Italian evidently illustrate a case o late semantic merger between the two words.

Historically speaking, this is a parallel innovation in these two languages, whether

due to areal or to typological convergence.

Conversely, the colexication hear eel ound in Mwotlap is also attested in all

known languages o Vanuatu, and was demonstrably inherited rom a similar pat-

tern in their common proto-language: Proto North-Central Vanuatu *roo hear,

smell, eel (Clark n.d.), rom Proto Oceanic *roor.

Finally, the presence o exactly the same colexication in Bislama, the pidgin/creole

o Vanuatu, historically results rom language contact. Te verb harem hear, smell,

eel, despite refecting English hear him in its orm, borrows its s emantics directly

rom the lexical structures o Oceanic languages, the vernacular substrate o Bis-

lama (Camden 1979: 5556).

In sum, colexication may result historically rom typological convergence, rom

genetic inheritance, or rom contact-induced change just like any other structural

eature o a language.

3. ypological prospectsTe observation o colexication does not only provide insights on individual lan-

guages or language groups. In theory, one can also conceive the possibility o ormu-

lating typological hypotheses in this domain, just like in other domains o languageresearch. Te ollowing paragraphs attempt not to state actual acts most examples in

this section being hypothetical but to dene the orm that uture research will be able

. See Browns maps on the colexication o hand/arm, nger/hand, as well as Kay

& Mas on green/blue or red/yellow, in the World Atlas o Language Structures by

Haspelmath et al. (2005).


to give to the ormulation o universals, whether absolute or implicational, regarding

the typology o colexication.

An absolute universal would take the orm (2), or its shorter equivalent (2):

(2) I alanguage lexiessenses1withtheormX,thensenses2willbelexied

inthesameway.

(2) Alllanguages colexiy thepairo senses s1s2.

An example o this o course subject to empirical check could be the pair male

sh vs. emale sh, which is apparently never ormally split in the worlds lexicons

as opposed to mammals, or which separate lexication is common. Interestingly,

this theoretical case should normally not come up in the data, given the method cho-

sen to distinguish between senses in the rst place (3.1.): the condition was to retain

only those sense distinctions that are attested in at least one language. Yet a lighter

version o (2), in terms o statistical tendencies (Most languages colexiy)

would be perectly acceptable, as would its symmetrical counterpart (Very ew lan-

guages colexiy ).

As or implicational universals, they can associate two cases o colexication:

(3) I alanguage colexiess1ands2,thenitwillalsocolexiy s3ands4.

For example, a likely assumption would suggest that i a language colexies arm andhand, then it will do the same or leg and oot.8 Or i it colexies paternal uncle

and maternal uncle, then it will also colexiypaternal aunt and maternal aunt

A subtype o this ormula would be (3):

(3) I alanguage colexiess1ands2,thenitwillalsocolexiy s2ands3.

For example, i a language colexies upper arm and hand, then it will probably

colexiyorearm and hand too. Or, i word and language are colexied, then

speech should be able to take the same orm. As these (ctitious) examples sug-

gest, this sort o ormula typically applies when the three senses can be conceived as

showing some orm o logical, cognitive ordering, so that s3 typically comes

between s1 and s2: e.g., because the orearm is physically located between the upper

arm and the hand, the colexication o the latter two makes it likely that the item inthe middle should be lexied identically. Te case oword< speech < language which I

intuitively suggest here would illustrate a similar, but more gurative, case o ontologi-

cal hierarchy between reerents.

. Counterexamples to this potential universal can however be ound, such as Lo-oga (orres

Is, Vanuatu, Oceanic group; pers. data), where arm = hand, but legoot.


10/29

2nd proofs

1 Alexandre Franois

Another variant o this ormula would be (4):

(4) I alanguage colexiess1ands2, thenitwillnot colexiy s2ands3.

whichmaybealsoormulated as:

(4) Althoughtheremayexistlanguages thatcolexiy s1ands2, andothers

thatcolexiy s2ands3, nolanguage colexiestogether thethreesenses

s1s2s3.

One possible example o such a ormula would be: i a language colexies person

and male person, then it will not colexiy that is, it will treat distinctively male

person and husband. Indeed, one can think o many languages where person and

male person go together (as with French homme), and many languages where male

person is the same as husband (as with Latin vir), but none until urther research

is done where the t hree are lexied the same.

Another kind o implicational universal would associate colexication with a cri-

terion outside the lexicon, as in (5)(5):

(5) I alanguage colexiess1ands2, thenitwillhavethelinguistic propertyP.

(5) I alanguage hasthelinguistic property P,thenitwillcolexiy s1ands2.

An example o (5) could be a statement about parts o speech, such as: i a language

colexies black and darken, then it treats adjectives as (a subclass o ) verbs. A pos-sible illustration o (5) would be something like: i a language doesnt distinguish

count nouns rom mass nouns, then it will colexiywood and tree. Needless to say,

all these examples are intuitive, and would only make sense i conrmed by relevant

empirical data.

Finally, one could conceive possible correlations between certain instances o

lexication and specic properties o the languages environment. Tis would lead to

universals or at least tendencies such as:

(6) I alanguage colexiess1ands2, thenitsenvironmentwillhavethe

property P.

(6) I alanguagesenvironmenthasthepropertyP,thenthislanguage will

colexiys1ands2.

Te term environment, used in a unctional perspective, encompasses all propertiesthat are not strictly linguistic, but which are somehow associated with the language or

its speaking community. One could thus imagine the ollowing sort o hypothesis: i a

language colexies cow and bull under a single term, it is likely that this language is

used in a society where this particular gender dierence is unctionally less relevant

that is, where cattle arming is not practiced traditionally. Similar types o correla-

tion were mentioned in 3.4. above, with Browns cultural-cognitive interpretations o

certain cases o colexication.


. Lexical semantic maps

So ar, the method here exposed has entailed the identication and manipulation o

senses, in the orm o unordered lists. Te only visual representation proposed (Figure 1)

took the simple orm o overlapping sets, still with no specic internal organization.

Yet the high number o senses involved, combined with the number o languages

potentially explored, would ideally require dening a more sophisticated way o orga-

nizing and presenting the results o our semantic observations. Tis is what I will now

propose to do, in a orm suggested by current research in grammar typology, 9 and

systematized by Haspelmath (2003): semantic maps.

.1 General principles o semantic mapsHere is how Haspelmath (2003: 213) denes semantic maps: A semantic map is a geo-

metrical representation o unctions in conceptual/semantic space that are linked by

connecting lines and thus constitute a network.

Essentially speaking, a semantic map takes the orm o a two-dimensional chart,

and represents a selection o meanings (senses in my terminology, unctions in

Haspelmaths). Tese meanings are ordered in space according to certain principles,

and explicitly interconnected, thus orming a semantic network. In itsel, this semantic

map constitutes an etic grid which claims to be language-independent, a coherent

chunk o a universal network. Tis universal grid t hen serves to visualize the emic

categorizations which are made by each specic language: or a given orm in a given

language usually understood in synchronical terms it then becomes possible to

identiy, on the universal map, those meanings that are covered by this orm, and those

that all without its scope.10

Te whole methodology presented by Haspelmath is compatible with the model

o lexical typology which I here propose to develop. Te only dierence is that he

explicitly designs his model as a way to represent the geometry ogrammaticalmean-

ing, while the present discussion deals with the lexicon. Yet, even i all his examples are

taken rom acts o grammar, he himsel suggests that his model should theoretically

be compatible with the lexical domain too (2003: 237). In a way, the ollowing pages

may be seen as an attempt to apply to the lexicon the principles dened by Haspelmathor drawing semantic maps.

. See, or example, Anderson (1982) or the perect; Cro et al . (1987) or the middle voice;

Jurasky (1996) or the diminutive; etc.

1. For a visual illustration o this principle, see Figure 4, and the gures inAppendix 3.


11/29

2nd proofs

1 Alexandre Franois

O course, several authors have already proposed to represent lexical semantics, and

in particular polysemous networks, in the visual orm o a diagram or map. But most

oen, their intention was to illustrate a pattern o polysemy specic to one language, or

one group o languages.11 By contrast, the maps I propose to draw here claim to have a

universal value, that is, to provide results that are virtually independent rom any par-

ticular group o languages. Tis is coherent with the stance taken by Haspelmath or

his grammatical maps: Te conguration o unctions shown by t he map is claimed

to be universal (2003: 217). O course, the quality and precision o a map will depend

on the number and genetic diversity o the languages observed. But essentially, what-

ever result comes out o such a study, should be able to claim universal relevance. An

important consequence o this principle is that any new data rom a natural language

should thereore be able to alsiy these results. As Haspelmath (2003: 232) puts it,

Every semantic map can be interpreted as making a universal claim about languages,

that can be alsied easily.

.2 Connecting senses togetherIn comparison with the simple ormat o a sense list, the main interest o semantic

maps is to organize the polysemous network in a way that makes explicit t he various

semantic connections between these senses. Tis is shown visually, on the one hand,

by the iconic grouping o close senses in contiguous areas o the map; and on the otherhand, by the use o explicit connecting lines to visualize semantic paths.

Judgments o closeness between senses are established in a dual ashion: rst, by

taking into account the ontological properties o each sense; second, by examining

empirical data rom various languages.

Te intrinsic ontological properties o each sense can legitimately be taken into

account in order to suggest a semantically plausible ordering between senses. For

example, suppose one came across an array o senses such as the one observed with

the verb lin Mwotlap (Franois, in prep.):12creak; name a child so-and-so; crow;

11. Tus, the maps ound in Matiso (1978) intend to represent certain semantic associations

specic to the ibeto-Burman amily; those in Evans (1992) or Evans & Wilkins (2000: 560)

apply to Australian languages; Eneld (2003) to Southeast Asian languages; yler & Evans (2003[2001]: 125) propose a semantic network specic to the polysemy o English over Some proj-

ects aim at representing semantic associations at the level o the whole lexicon, but they are

still, by denition, restricted to a single language c. Gaume et al. (this volume) or French; or

soware such as Tinkmaps Visual Tesaurus or English. (www.visualthesaurus.com)

12. Even though this list o senses, as well as its representation in Figure 2, are drawn aer

the polysemy o just one word in one language, I propose that it is ctitiously understood, or

the purpose o this demonstration, as i resulting rom cross-linguistic comparison. Indeed the

orms o reasoning that apply in both cases whether we consider one polysemous network,


mention s.o.s name; yell; invoke a divinity; scream; bark; hail s.o.; call s.th.

such-and-such Tis kind o simple list, presented in random order, makes it hard

to identiy the semantic links between these senses. But unctional considerations

allow certain senses to be grouped according to their common semantic properties.

Tus, several senses reer to the emission o intense high-pitched sounds, whether

by humans (yell, scream), animals (crow, bark ) or objects (creak). Other

senses reer to human social activities that consist in uttering the name o another

person; this can be done or the purpose o calling out to someone (hail s.o., invoke

a divinity), or or the purpose o reerring to them (mention s.o.). Finally, the act o

uttering a name may reer to the s ocial act o giving a name to someone typically a

child or to something.

Semantic connections can then be proposed, which chain senses according to

their unctional similarities. Tese connections may then easily be represented in

space, in the orm a visual graph such as Figure 2:

(animal)cry

(cock)crow

(dog)bark

(s.o.)scream

(thing)creak

(s.o.)yell

(s.o.)hail s.o.

call s.o.sname

mention s.o.

name a childso-and-so

invokedivinity

name s.th.so-and-so

Figure 2. Senses may be linked based on unctional properties.

Crucially, because the semantic connections here proposed are supposedly based on

ontologicalproperties o the notions reerred to, this means they must normally be

conceived as independent o any specic language. Tat is, even though the list o

senses itsel was initially based on the observation o actual languages, ultimately the

act that a sense s2 will be understood as orming the missing semantic link between

s1 and s3, should not depend on any particular language, but simply on t he intrinsic

properties o each sense. For example, the act o hailing someone (by shouting theirname) constitutes a logical t ransition between shouting (in general) and uttering

s.o.s name. Tis organization o meaning must be understood as driven not by idio-

syncrasies o any specic language, but rather by universal characteristics o the real

or the intertwining o several such networks into one are undamentally the same, at this

particular stage o the study.


12/29

2nd proofs

1 Alexandre Franois

world or more exactly, o the world as it is perceived by the human brain and ltered

by human activities.

Tis being said, it remains obvious that the connections proposed between each

two senses, and more generally the semantic map that results rom these connections,

cannot be directly observed in the material world, and thus constitute hypotheses on

part o the observer. Tis means that they must be amenable to proo or demonstra-

tion, that they are open to debate, and that they should be alsiable. Te problem

is, at least some o these semantic hypotheses about how two senses should be

connected may ultimately depend on the linguists intuition. Tis is potentially an

issue (see also the discussion in 3.1.), because the conscious representations o the

world by an observer do not necessarily match the subconscious connections which

are actually made by the speakers brain. It is t hereore necessary to dene a method o

alsication that would rest on empirical observation.

Te method suggested by Haspelmath indeed resorts to observable data rom

actual languages. Te basic idea is that senses should be arranged in space in such a

way that each lexical unit in one language occupies a contiguous area on the seman-

tic map (2003: 216). Furthermore, each specic connecting line should refect the

existence o at least one attested case o a direct lexical connection between these two

senses, in any o the worlds languages. Tus, supposing one language were ound t hat

only colexied a sense si and a distant sense sj but none o the other senses tentativelyproposed in-between, then the background map should be redesigned, and a short-

cut connecting line added between thes e two senses. Conversely, i all words colexiy-

ing si

and sj

also include, in their polysemy, the various intermediate steps proposed

along the unctionally-based semantic chain, then the hypothetical map can be s aid to

be conrmed by empirical data.

Incidentally, it may happen, on some occasions, that two distinct paths may be

dened in order to relate two senses on the map, with no strong reason or choosing

between these two paths. We shall see precisely an example o this in 5.3.4., where two

dierent semantic hypotheses will be shown to equally account or the colexication o

breath and supernatural power. Insoar as this sort o hypothesis is also supported

by empirical data in this case, the existence o two distinct sense chains attested in the

worlds languages nothing prevents us rom representing this double path on the map.

.3 Choosing a pivot notionIn section 3 above, I briefy mentioned the necessity to choose a specic notion (sense)

as the pivot o the map. Tis requires justication, especially because this principle

seems to dier rom Haspelmaths (2003: 232) method or drawing grammatical maps.

Choosing a specic sense (e.g., hail s.o.) as the pivot entails that the empirical

data to be observed must consist exclusively o lexical units that specically include


this sense in their polysemy. Tis important requirement is a precaution against the

risk o starting an open-ended map with ever-shiing boundaries. For example, con-

sider the colexication ohail and (animal) cry. Ihail were not given any special

status, nothing would then prevent us rom including in the data words that encom-

pass (animal) cry as one o their senses, yet having no connection with hail: or

example, a verb meaning (animal) cry; (s.o.) cry out in pain; weep (c. Eng. cry).

I this were allowed, then the map would extend so as to include all t he semantic con-

nections associated with the sense weep, and so on and so orth. Such a map with no

center would shi indenitely so as to gradually enclose the w hole lexicon. Despite

the immense interest o potentially achieving a map that would represent the global

geography o the human mind (Cro 2001), such a conguration would rapidly lead

to uncontrollable results that would raise obvious technical issues, and whose signi-

cance in terms o scientic inormation would end up being questionable. It is thereore

saer to circumscribe in advance the scope o the map that is to be drawn, by providing

one sense with the special status o pivot or centre. Incidentally, I propose to translate

typographically the special status o the pivot notion, by using small uppercase and

braces e.g., {hail} as opposed to the other senses o the network e.g., bark.

Tere is a corollary to this principle. I two senses s1 and s2 are attested to be

colexied in the worlds languages, the map centered on s1 will be a dierent map rom

the one centered on s2. Tus, the choice o {hail

} as the pivot will trigger a specicsemantic network one that can be called, in short, the lexical map o {hail} which

will tell a totally dierent story rom the choice o {(animal) cry}. Quite logically, how-

ever, one can predict that these two maps will have a whole chunk in common that

is, the connection between these two s enses, plus whatever urther senses are attested

to colexiy with these two senses together. Tus, supposing a language were ound

that colexied (animal) cry scream hail call, then each o the our semantic

maps centered on each o these senses would necessarily have to include this particular

chain o senses along with other ramications specic to each map.

Finally, note that the status o pivot o a lexical map has nothing to do with the

notion o prototype, which is only relevant to the description o individual lexemes.

Tus, it is perectly possible that a typological map centered on the sense {hail} incor-

porates a lexeme X whose polysemy encompasses only those senses that appear to

the le ohail in Figure 2 above (hail, scream, creak, (animal) cry ). Inthis particular language, it is likely that a prototype-based approach would describe

this word X as being built around the prototypical meaning shout with high-pitched

voice, scream; the sense hail would be nothing more than a peripheral oshoot o

that core meaning regardless whether or not it is t he pivot o the universal map that

includes it.

Another dierence is that the denition o a prototypical meaning, in the

(language-internal) description o a word, constitutes an interpretative claim about


13/29

2nd proofs


this word that may be challenged or alsied. On the contrary, the selection o a

given notion as the pivot o a (universal) lexical map entails no claim at all: it is sim-

ply an arbitrary choice, the starting point beore any lexical map may even begin to

be drawn.

. Elaborating a universal map or breathe

In order to illustrate in ull detail the typological method I am here advocating, I now

propose to delve into a specic notion, and build the lexical semantic map that will

best render the various polysemies associated with it in the worlds languages. Tis is

what I will do in the remainder o this article, around the notion breathe.

Te notion breathe is here understood as the physiological activity o breathing

characteristic o humans and animals. I will rst observe, or each language o the cor-

pus, the set o other senses w ith which this notion is colexied. Ten I will attempt to

draw the lexical map o the notion {breathe}. Te nal orm taken by these two steps

appear respectively as able 2 and Figure 5 in Appendix 2.

Tis small case study rests on a corpus o 16 lexical headwords in 13 genetically

diverse languages. Each entry consists o either a single word, or a lexical root, in which

case several words are encompassed under the same entry. In particular, it is requentthat the noun and the verb associated with t he notion {breathe} dier ormally rom

each other; in this case, I have organized arbitrarily the data in the appendices in such

a way that the deault headword is the noun, while the cognate verb, when ormally

dierent rom it, has a secondary status (loose colexication).13

Te lexical database presented in Appendix 1 shows a total o 114 words involved

in the comparison. O course, richer data, taken rom more languages, would logically

result in richer results, with e ven higher typological signicance. However, the corpus

here analyzed was judged at least sucient or the purpose o illustrating the typologi-

cal method here proposed.

.1 A rst overview o the verb breatheIn Makonde, a Bantu language o anzania, the verb ku-pumula colexies breatheand take a rest. Tis semantic connection has a transparent motivation. In the rst

place, the physiological act o breathing becomes particularly signicant cognitively

salient aer one has held his breath while making a physical eort. Te act o sitting

13. Tis is whyable 2 shows plain + signs in the rows act o breathing and pu o breath,

but bracketed [+] signs, standing or loose colexication, in the rst row breathe. See also the

isolectic sets in the maps oAppendix 3.


down or a minute aer an intense eort, or even o resting or a whole day aer a week

o work, can be seen as a semantic expansion o this initial meaning, even when what

is relevant is not so much the act o breathingper se, as that o ceasing an eort. I we

add pause or breath as the missing semantic link (3.4.) between these two senses, the

polysemy oku-pumula can be represented using a string o three senses breathe

pause or breath take a rest. Tis is a classical case o colexication originating in

semantic extension.

Makonde is not the only language to have developed this polysemy. English showsa case o loose colexication (3.2.) between breathe and take a breather. Te colexica-

tion, whether strict or loose, obreathe and take a rest is also attested in Sar (noun

koo), in Arabic (root r.w.), in Nahuatl (verb imiiyo), in Mwotlap (verb mkheg), in

Nlmwa (root hor-), in Russian (root *du[x]); but not in Latin, Greek or Inuit. Tis

is enough evidence to propose this case o colexication as typologically signicant.

Interestingly, Mwotlap mkhegcan also equally be us ed or any period o rest, i.e.,

not only minutes o pause within hours o work, but also days o pause within months

o work that is, what we would call take a vacation. Since certain languages do not

go that ar in the semantic expansion o {breathe} (e.g., English would hardly describe

a month-long holiday as take a breather), it is wiser to dene ormally not three but

our dierent senses here: breathe, pause or breath, take a rest and take a vaca-

tion. Out o these our senses, we will say that English colexies only three, whereasMwotlap covers them all. Incidentally, this proposal does not involve the claim that

these senses are necessarily distinct or the Mwotlap speaker and it is perectly likely

that take a rest and take a vacation should be grouped together under an emic

approach. But what is relevant here, or the specic purpose o language comparison,

is that these two unctional situations are colexied in Mwotlap, but distinguished in

English; hence the choice to treat them, in an et ic perspective, as i they were distinct

semantic units (see discussion in 3.1.). Incidentally, Russian otdyx, etymologically

connected with dyatbreathe, means both rest and vacation.

In a similar way, the Nlmwa verb horn has added an extension to the mean-

ing take a rest, namelystop doing s.th., cease (e.g., Co horn o khiiboxa pwaxim

tavia Stop beating your dog!). Tis semantic oshoot clearly adds a new sense to the

potential polysemy o {breathe}.

Te same observations can be made or other senses related to { breathe}. Insome languages (e.g., Greekpne), the same verb is used or breathe, or blow (i.e.,

a person blowing actively into s.th., like a fute) and/or or (wind) blow. A urther

connection that is sometimes attested is between blow and whisper, with a shi

towards the notion o articulated speech. Tus in Araki (Franois 2002), the verb sono

connects the notions blow, pu, blow into s.th. and talk, tell a story see also

the derived noun sonosono speech, story; language. Likewise, the French verb souer

means both blow, pu and whisper, prompt.


14/29

2nd proofs

1 Alexandre Franois

But the latter example o colexication potentially raises an issue, because it

involves the sense blow rather than breathe (breathe is mapu in Araki, respirerin

French). Consequently, it should be kept aside rom the semantic network o {breathe}

strictly speaking, to avoid the risk o shiing the center o observation rom one sense

to the other, and thus expanding innitely each polysemous network (see 4.3.). In

other words, the evidence so ar allows us to include blow among the senses directly

connected to {breathe}, and utter among the senses directly connected to {blow};

but it does not illustrate any colexication between {breathe} and utter.O course, the conclusion would be dierent i we came across languages that did

witness the colexication o these two senses. Tis is in act the case with the noun

hor- in Nlmwa, which means both breath, breathing and spoken message.

Te English phrase I wont breathe a word also illustrates the potential connection

between breathe and utter, arguably via a missing linkwhisper (as in breathe a

prayer). Tese two examples nally legitimize the inclusion outter, speak in the map

o {breathe}.

.2 From the sense list to the mapBeore going any urther, it may be useul to recapitulate our rst ndings in a visual

orm. A simple way to do so would be to draw a table, based on the list o senses that

have been observed to potentially colexiy with the pivot notion {breathe}. Each col-umn corresponds to one o the languages I have been reviewing so ar, representing

a subset o my corpus. Tis leads to able 1, a partial representation o the sense list

under construction here (see Appendix 2 or the complete table).14

Table 1. Examples o colexication associated with {breathe}

english

breatherussian

du[x]mwotlap

mkhegnlmwa

hornaraki

sonofrench

soufer

breathe + + + +

take a rest [+] [+] + + +

be on vacation [+] +

cease to do +

(wind) blow + [+] +

(s.o.) blow + [+] + + +whisper + + +

utter, speak + [+] + +

1. Te typographical contrast between plain plus + and bracketed plus [+] corresponds

respectively to strict and loose colexication (see 3.2.).


Note that Araki sono and French souerare included here or the sake o cross-

linguistic comparison. However, as discussed above, they cannot take part in the

corpus, because the sense {breathe} chosen as this studys pivot (rst row) does not

belong to t heir polysemy.

Tis representation in the orm o a table has the advantage o being clear and

straightorward. Yet, it has t he drawback o treating all senses on the same level. It may

be more interesting to underline the semantic links that relate certain senses with oth-

ers, and which orm unctional subsets within the network (s ee 4.2.). For example, wehave seen that the sense be on vacation is a semantic extension o the sense take a

rest, itsel being closer to the more literal meaning pause or breath; and that cease

to do is another, independent oshoot omake a pause. Te chain breathe pause

or breath take a rest be on vacation thus has a coherence o its own, which is

clearly distinct rom the chain breathe blow whisper speak.

A more inormative and graphic representation would thus take the orm o

a semantic map, a diagram showing all the senses attested, together with the most

likely semantic connections that link them. Tese connections are rst based on

intrinsic semantic properties, and are then checked against empirical data (see 4.2.).15

Tis brings about the tentative map o Figure 3.

pause for breath

take a rest

take a vacation

(s.o.) blow

(wind) blow

whisper

utter, speak

cease to do

Figure 3. A rst semantic map or {breathe}.

1. o be precise, the polysemy o Nlmwa hor- breath, breathing; spoken message raises

an issue, because it does not include the senses ( blow and whisper) which unctional consid-

erations suggest to posit as intermediate between breathe and speak (whisper in Nlmwa

is nyomamat). In theory, a rigorous application o the principles exposed in 4.2. should trigger a

shortcut line between these two senses. However, the strong unctional motivation owhisper

as a likely missing link, and the act that the whole chain is empirically attested in other lan-

guages, suggests we may be dealing with a case I have not discussed yet: that is, the possibility

that an initial chain o senses s1s2s3s4 may have evolved historically so that some interme-

diate links got lost via lexical replacement and only s1 and s4 remained colexied. Although

this is debatable, I choose to inringe the rule here, and to keep on the map the intermediate

steps o the path, based on unctional motivations. Tis is why the Nlmwa set appears as non-

contiguous in Figure 4, in spite o the ideal design o semantic maps in Haspelmaths terms.


15/29

2nd proofs

1 Alexandre Franois

Once it is established albeit incompletely a semantic map like Figure 3 consti-

tutes a universal etic grid against which emic categories o specic languages may be

described. Each lexical headword (word or root) selects a particular subset out o the

total range o potential senses. Tis is made clear by Figure 4, which converts the data

o able 1 into graphic sets. By analogy with the concept o isoglosses, I propose to call

these sets isolectic sets.

pause for breath

take a rest

take a vacation

(s.o.) blow

(wind) blow

whisper

utter, speak

cease to doEnglish breathe

Mwotlapmkheg

Nlmwa hor-

Russian *du[x]-

Figure 4. Some isolectic sets around the notion {breathe}.

Te most instructive point here, in terms o typology, is that the array o cross-

linguistic variation, ar rom being innite and random, appears to be relatively lim-

ited. O course, the more languages are considered, the more senses will appear in

the chart. But even at the small scale o these rst observations, the act that the

same patterns o polysemy recur again and again across language amilies is, in itsel,

o considerable interest in the search or potential language universals. Tis sort

o cross-linguistic comparison can help see which patterns o polysemy are typo-

logically more common than others (see 3.4.): or example, while the our languages

presented here all share the colexication obreathe with take a rest, only one has

gone as ar as to include the meaning cease to do. O course this result with only

our languages is not signicant; but the possibility o extending the observation to

virtually hundreds o languages suggests t he sort o research that may be carried out

in the uture.

.3 Exploring the noun breathTe preceding paragraphs have presented the principal cases o colexication associ-

ated with the verb breathe in my corpus. A much richer semantic network arises i

one addresses the domain o nouns. Many languages possess a noun which is cognate

with the verb breathe (Eng. breath) I will call it here the {breathe} noun.


.3.1 Breath, breath o air, scent In some languages, as one would expect, this deverbal noun carries with it part o the

polysemy o the verb breathe; but most oen, languages provide that noun with its

own polysemy, which warrantsa specic description. Tus, to take the case o Mwot-

lap, the verb mkhegbreathe; take a rest; be on vacation has a directly derived noun

n-mkheg, which means equally breathing, rest and vacation. But it also pos-

sesses a cognate noun n-mkhe with its own particular semantics: breath, smell,

breath o lie, etc.

Te literal meaning o the {breathe} noun is normally to designate the physical

activity, or manner, o breathing (Eng. pause or breath; be short o breath). In some

languages, it also expresses the portion o air inhaled or exhaled during the act o

breathing, including its physical properties such as temperature or smell (hot breath;

bad breath). By extension, the same word is sometimes used or all sorts o smells,

even when unrelated to an actual process o human breathing: e.g., Mwotlap n-mkhe

ttnge the scent (lit. the breath) o fowers. Trough a similar shi between man and

nature, the human activity o breathing is sometimes colexied with natural phenom-

ena involving motion o air, such as breath o air, wind or even cold air.

Tese dierent senses seem to be articulated into two chains: on the one hand, a

chain human act o breathing air in motion: breath o air wind cold air; on

the other hand, a chain human act o breathing air coming rom human mouth smell coming rom human mouth smell, scent in general. o take just one exam-

ple, Latin spritus, derived rom spro breathe, is attested with all these meanings

(except or cold air).

.3.2 Lie, spirit, mind, eelings But probably the most signicant polysemy that is attested with { breathe} nouns is

the lexical eld o lie and soul. Tis time, among the various properties associated

with the act o breathing, the one which is most relevant here is a universal physiologi-

cal observation: namely, that the phenomenon o breathing is the most salient property

that distinguishes a live creature rom a dead body.

Tus, {breathe} nouns or verbs are requently perhaps universally attested in

phrases related to t he semantic notions o lie and death: see Eng. breath o lie; draw

ones last breath; breathe lie into s.th. In Latin, the verb exspro(rom spro breathe)means literally breathe s.th. out, but also serves as a euphemism or breathe ones

last, die (> Eng. expire). Russian iz-dyxatdie, etymologically connected to the root

du[x], is exactly parallel to Latin ex-spro.16

1. Te reason why the sense die is not represented on the nal semantic map o {breathe}

(Appendix 2) is because this meaning is always obtained indirectly, through lexical or


16/29

2nd proofs

1 Alexandre Franois

Tis is how certain languages have come to colexiybreath and lie. More pre-

cisely, the {breathe} noun is oen related, whether historically or synchronically, with

a word whose meaning could be described as the principle o lie, insoar as it can be

conceived as specic o an individual. Indeed, while these languages oen possess a

separate noun or the abstract concept lie (Greekbios, Lat. vta, Arabic aia ),

they also oen make use o another term when it comes to embodying this abstract

principle, as it were, into an individual being. Tis is how many i not all cultures

around the world have elaborated the non-trivial notion o the soulor spirit: that is, thevital orce o an individual, insoar as it is opposed to the inert body.

Needless to say, a wide variety o conceptions can be carried by this notion o

spirit, depending on cultures, religions, times and people. Despite the risk o simpli-

cation, this diversity can perhaps be reduced to a ew prototypical concepts. At least,

I shall mention here those concepts that are lexied, among the worlds languages, in

direct connection with the notion {breathe}.

In some languages, the {breathe} noun embraces the psychological activity o

an individual, in its various maniestations. For example, Classical Latin animus17 is

attested with the ollowing meanings: vital principle o an individual: soul; seat o

reason and intelligence: mind; seat o will and desire: will; seat o eelings and pas-

sions: heart; seat o courage and vital energy; strong passions: pride

Te semantic range is not necessarily as wide as t his, and is sometimes restricted tojust a certain type o eeling. o take another Latin example, the noun spritus, besides

its other meanings mentioned in 5.3.1., is also attested with psychological senses; but

as ar as Classical Latin is concerned, these are essentially restricted to pride, arro-

gance, sel-importance. During the later history o L atin and o Romance languages,

the set o psychological meanings related to spritus has enriched considerably. Tus,

French esprithas a wide polysemy o its own, which includes mind, thought, intelli-

gence, wit, seat o eelings, character, moral disposition, rame o mind, mood.

A ew phrases illustrate these senses, such as garder lespritkeep in mind, avoir

lesprit vihave a quick mind, avoir de lespritto be witty, avoir lesprit rire to be in

phraseological derivation, but never directly (strict colexication). For obvious reasons, no

language is ound where die and breathe are expressed by exactly the same orm in syn-chrony. As a principle, those senses which are attested nowhere in strict colexication with the

pivot notion do not qualiy or inclusion in its semantic map (see 3.3.).

1. Admittedly, animus did not have breathe nor breath among its senses in the synchrony

o Classical Latin. However, it is etymologically linked to Greek anemos wind and Sanskrit

aniti breathes; and more importantly, it is closely cognate with the noun anima, whose wide

polysemy does include breath and wind. As a result, I take anima as the relevant headword

or Latin (see 7.1.3.); animus is only included in the corpus by virtue o its synchronic cognacy

with anima (loose colexication).


a mood or laughing, dans lesprit de lpoque in the spirit o the age, esprit dquipe

team spirit, retrouver ses esprits to collect ones wits Incidentally, because French

esprit like Eng. spirit no longer shows any connection with {breathe} in syn-

chrony, it can only be included in our corpus on a historical basis. In case we want

to restrict our observations to synchronical polysemies, then the examination o this

root should be restricted to Classical Latin spritus, whose semantic array is already

wide (7.1.3.).

Similar semantic extensions can be ound in other languages, including in theorm o synchronically coexisting senses. For example, it is remarkable that Standard

Arabic also translates some o the psychological senses o Fr. espritwith ru a noun

related to {breathe} (7.1.10.): e.g., ru al-taaun team spirit, al-ru al-arbiya

warlike spirit. Te polysemy o Russian duxalso presents similar characteristics in

synchrony even more i one considers the whole set o words that orm the cognate

set o the root *du[x]18 (7.1.4.).

.3.3 Soul, spirit, supernatural being Te group o senses just reviewed (spirit, mind, character ) orms a branch o

its own in the semantic map o { breathe}, covering the domain o psychological and

mental qualities o the socialized person. It should be careully distinguished rom

another concept: the soul. Te semantic nuance is amiliar to all Latinists, since it is or-mally distinguished in Latin as (masculine) animus vs. (eminine) anima. While animus

describes the various aculties, eelings and emotions o individuals in their social activ-

ities, anima has a deeper existential meaning, as it reers to the primal aculty o being

alive see also the derived noun animalliving being. Tereore, nouns like anima

will be typically used in contexts dealing not with social behaviour, but with death. In

this perspective, the soul can be described as that part o an individual which leaves

the body when death comes. Depending on the cultural context, this separation rom

the body will be understood either as the complete disappearance o the soul, or, on the

contrary, as its survival in dierent orms: migration o the soul to an invisible abode o

the dead, restless wandering as a ghost in the present world, reincarnation (metempsy-

chosis) into a new human body, or metamorphosis into a supernatural being.

One may think that these cultural issues are not relevant or our linguistic study,

but they are. Only the understanding o such religious belies makes it possible todene a satisactory semantic path between, on one end o the semantic chain, the

notion o breathing, and on the other end, the representation o ghosts and other super-

natural beings, whether in an animist or a monotheist context. Tis polysemy can be

1. Amongst the various lexical items that are etymologically related with this root, the noun

dua soul, spirit has received special attention in Wierzbicka (1992: 31.).


17/29

2nd proofs

1 Alexandre Franois

illustrated again with Fr.esprit(< spritus): besides the mental and moral senses used in

a social context (taking over the semantics oanimus), it can also reer to the soul o a

living being (anima), including in the orm o a ghost (e.g., croire aux esprits believe in

ghosts). Finally, espritcan designate any supernatural being o divine nature (lesprit

du euve the spirit o the river), whether good (esprits clestes heavenly spirits) or

evil (esprit malin evil spirit); and in the context o a monotheist religion, the same

word may even come close to reerring to the supreme divinitypar excellence, as in

le Saint-Espritthe Holy Spirit.Tis impressive range o spiritual meanings is not exclusive to the lexicon o Latin

(animus, anima, spritus) and o its daughter languages. Surprisingly similar patterns

o polysemy are ound elsewhere: Greekpskh andpneuma; Sanskrit tman; Russian

du[x]; Arabic r and nas; Aleut anri; Nahuatl imiiyo; and so orth (see Appendix 1).

.3. Going rom breath to supreme spirito be precise, there are two ways one could account or the inclusion o supernatural

beings in the semantic map o {breathe}. One hypothesis would involve a generaliza-

tion process, whereby the soul o a human individual, insoar as it is said to survive

aer death in the supernatural orm o a ghost, would serve as a model or all other

supernatural creatures, even when they do not originate in a deceased person. In this

case, the likeliest semantic chain would be:

breath (breath o) lie vital orce o an individual, s.o.s spirit immaterial

part o an individual that survives death: soul s.o.s ghost supernatural

being, even when not o human origin; a spirit, good or e vil

Te likeliness o this scenario is conrmed by the existence o similar semantic

shis with other lexemes, though unrelated to {breathe}. For example, the Mwotlap

noun na-tmat(Franois, in prep.), etymologically dead person, is a polysemous word

that colexies deceased person wandering soul o a deceased person, ghost

monster, spirit; any supernatural being, whether good or bad theBiblical Devil.

A second hypothesis would make a shortcut between t he very act o breathing and

the notion o divinity, with no need to posit soul, ghost as a missing link. Indeed, in

many cultures, the immateriality o divine entities is metaphorically compared with aninvisible breath o air, a magic wind. Tis divine wind may s ometimes be blown into

a thing or a person to endow it with holiness or supernatural power. Tis metaphor,

or example, underlies the use o Eng. inspiration(or an artist, a poet, a prophet) rom

Latin inspro blow into. Likewise, the Classical Greekpneuma, literally breath, breath

o air is attested with the meaning divine breath, 19 but never with the sense soul

1. Historically speaking, this specic sense, despite being already attested in Platos works,


or ghost. Finally, a process o metonymy triggers the shi rom divine breath to

the divine entityorsupernatural being rom whom a divine breath emanates. Tese

examples would thereore rather advocate or a second semantic path:

breath o air emanating rom a human person divine breath: supernatural

power emanating rom an immaterial entity supernatural being exhaling

divine breath, divine spirit

Because both chains seem to be semantically likely and empirically grounded, I preernot to choose between them (s ee discussion in 4.2.). Such ambiguity is not necessarily

an issue, and may well depend on the specics o each language or culture. It can be

easily represented on the typological map o {breathe} by drawing two distinct paths

leading rom breathing to supernatural being: see Figure 5 in Appendix 2.

.3. From soul to reexive markingFinally, a urther extension rom the sense soul, spirit is the designation o an

individuals person, essence or ego what one may dene as ones inner, deeper

identity, as opposed, or example, to ones social representation. Tis more or less

corresponds to the semantics o English sel.

Even more interestingly, this quite abstract meaning has sometimes grammatical-

ized into a refexive marker, in a way precisely parallel to English (know) your sel>

(know)yoursel. Tis semantic path is witnessed in three languages in my corpus. In

Sanskrit (7.1.1.), the amous concept tman (etymologically breathing, rom an-

breathe) has a wide semantic array, going rom breath o lie to vital orce, soul

and the sel, the abstract person as well as essence, peculiarity (o something). But

one o its principal uses in texts s eems to be as a grammatical marker or refexive; this

is especially clear rom the list o dozens o compounds based on tma- (o which only

a short selection is given in the Appendix), e.g., tma-ja knowing ones sel , tm-

vara master o ones sel , tma-ghta suicide, tma-grhinselsh

Likewise, the ordinary refexive marker or Standard Arabic is nas-(1sg possessed

orm onas, parallel to Eng. my-sel). Tis is in act a noun nas meaning soul,

essence, being, abstract person, sel, mind, psyche as well as the same And

crucially, this whole semantic array is closely connected via loose colexication

with the noun naas breathing, breath, breath o lie (root n..s). Tis example

was later spread by the Septuagint in their translation o the Bible. Whereas the noun pskh

soul, spirit had lost its etymological relationship to breath, the noun pneuma was still syn-

chronically the word or breath, blow o air: this is probably why it was chosen to translate

Hebrewruach breath, air; strength; wind; spirit; courage; temper; Spirit (Vine 1985: 240; see

the cognate Arabic ru in the appendix). Exactly in the same way, the semantic calque took place

in Latin with spritus breath, blow o air; soul rather than animus, because the connection o

the latter noun with wind, breath was then no longer perceptible.


18/29

2nd proofs


conrms the relevance o a semantic chain breathing breath o lie vital orce

person, sel reexive.

Te other Arabic root with a similar polysemy, r.w. apparently does not go tha

From Polyseme to Ling ChangeFrancoisSemanticMaps

Documents

Transcript of From Polyseme to Ling ChangeFrancoisSemanticMaps