MODALITY IN GRAMMAR AND DISCOURSE

TYPOLOGICAL STUDIES IN LANGUAGE (TSL)

A companion series to the journal "STUDIES IN LANGUAGE"

Honorary Editor: Joseph H. Greenberg

General Editor: T. Givón
Associate General Editor: Michael Noonan
Assistant Editors: Spike Gildea, Suzanne Kemmer

Editorial Board:
Wallace Chafe (Santa Barbara) Ronald Langacker (San Diego)
Bernard Comrie (Los Angeles) Charles Li (Santa Barbara)
R.M.W. Dixon (Canberra) Andrew Pawley (Canberra)
Matthew Dryer (Buffalo) Doris Payne (Oregon)
John Haiman (St Paul) Frans Plank (Konstanz)
Kenneth Hale (Cambridge, Mass.) Jerrold Sadock (Chicago)
Bemd Heine (Köln) Dan Slobin (Berkeley)
Paul Hopper (Pittsburgh) Sandra Thompson (Santa Barbara)
Andrej Kibrik (Moscow)

Volumes in this series will be functionally and typologically oriented, covering specific
topics in language by collecting together data from a wide variety of languages and language
typologies. The orientation of the volumes will be substantive rather than formal, with the
aim of investigating universals of human language via as broadly defined a data base as
possible, leaning toward cross-linguistic, diachronic, developmental and live-discourse data.
The series is, in spirit as well as in fact, a continuation of the tradition initiated by C. Li (Word
Order and Word Order Change, Subject and Topic, Mechanisms for Syntactic Change) and
continued by T. Givón (Discourse and Syntax) and P. Hopper (Tense-Aspect: Between
Semantics and Pragmatics).

Volume 32

Joan Bybee and Suzanne Fleischman (eds)

Modality in Grammar and Discourse

 MODALITY
IN GRAMMAR
AND DISCOURSE
Edited by

JOAN BYBEE
University of New Mexico
SUZANNE FLEISCHMAN
University of California, Berkeley

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
 The paper used in this publication meets the minimum requirements of
American National Standard for Information Sciences — Permanence of
Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Modality in grammar and discourse / edited by Joan Bybee, Suzanne Fleischman.
p. cm. — (Typological studies in language, ISSN 0167-7373; v. 32)
Papers from a symposium on Mood and Modality held at the University of New Mexico in
1992.
Includes bibliographical references and index.
1. Modality (Linguistics)--Congresses. 2. Grammar, Comparative and general—Mood-
Congresses. I. Bybee, Joan L. II. Fleischman, Suzarne. III. Series.
P299.M6M64 1995
415--dc20 95-17034
ISBN 90 272 2925 2 (hb.) / 90 272 2926 0 (pb.) (European; alk. paper) CIP
ISBN 1-55619-639-3 (hb.) / 1-55619-640-7 (pb.) (U.S.; alk. paper)
© Copyright 1995 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. • [Link] 75577 • 1070 AN Amsterdam • The Netherlands
John Benjamins North America • [Link] 27519 • Philadelphia PA 19118-0519 • USA
To the memory of Dwight Bolinger
 Contents

Introduction 1

I. Agent-Oriented and Epistemic Modality 15

Agent-Oriented vs. Epistemic Modality: Some Observations on
German Modals 17
Bernd Heine
The Expression of Root and Epistemic Possibility in English 55
Jennifer Coates
Contextuel Conditions for the Interpretation of poder and deber in
Spanish 67
Carmen Silva-Corvalán
The Obligation Modality in Western Nilotic Languages 107
Edith Bavin
The Gestural Expression of Modality in ASL 135
Sherman Wilcox and Phyllis Wilcox

II The Interactional Basis of Modality 163

The Development of Epistemic Sentence-ending Modal Forms and
Functions in Korean Children 165
Soonja Choi
The Interactional Basis of the Mandarin Modal néng 'can' 205
Jiansheng Guo
The Discourse and Interactive Functions of Obligation Expressions 239
John Myhill and Laura A. Smith
Apprehensional Epistemics 293
Frantisek Lichtenberk
Moods and MetaMessages: Alienation as a Mood 329
John Raiman
Viii Contents

III Irrealis Modality and Subjunctive 347

The Realis-Irrealis Distinction in Caddo, the Northern Iroquoian
Languages, and English 349
Wallace Chafe
On the Relativity of Irreality 367
Marianne Mithun
The Grammaticalization of Irrealis in Tok Pisin 389
Suzanne Romaine
The Evaluative Function of the Spanish Subjunctive 429
Patricia V Lunn

IV Modality and Other Categories of Grammar 451

Negation and the Modals of Possibility and Necessity 453
Frank Palmer
A Functional Theory of Complementizers 473
Zygmunt Frajzyngier
The Semantic Development of Past Tense Modals in English 503
John Bybee
Imperfective and Irrealis 519
Suzanne Fleischman

Subject index 553

Language index 565
Author index 569
 Modality in Grammar and Discourse
An Introductory Essay*

Joan Bybee
University of New Mexico
Suzanne Fleischman
University of California, Berkeley

This volume is the result of a symposium on Mood and Modality held at the
University of New Mexico in 1992, the goal of which was to bring together
linguists whose research has targeted this area of grammar but whose ap­
proaches to it reflect differing perspectives on functional linguistics. The
symposium was planned so as to include a diversity of languages, of foci
(synchronic and diachronic), and of theoretical orientations, especially with
regard to the interaction of morphosyntactic, semantic and discourse-prag­
matic factors.
Given the complexity of this linguistic domain, the many and diverse
ways it comes to be expressed in different languages, and the tendency of
researchers to work narrowly within the confines of their own theoretical and
methodological frameworks, there was reason to fear that symposium partici­
pants might not find adequate common ground nor a sufficiently common
metalanguage to be able to communicate with one another. However, just as a
similar symposium on tense and aspect a decade earlier (Hopper 1982)
confirmed the status of those categories as valid cross-language categories of
grammar—what we refer to, following Bybee & Dahl (1989), as 'gram
types'—, so too the current symposium succeeded in demonstrating that
despite differences in terminology, language areas, and theoretical perspec­
tives, we were in effect all examining similar phenomena and could partici­
pate in meaningful dialogue about our data and analyses. Thus while the
papers in this volume present a wide range of topics and perspectives, they
2 Modality in Grammar and Discourse

nonetheless converge around a number of key issues, and in the aggregate

seem to have succeeded in moving us toward a better understanding of the
functions of modality and its forms of expression in natural language.
In the course of the symposium several essential issues came up repeat­
edly regarding both the categories in question and the levels of linguistic
analysis at which they operate. One of the most basic of these issues concerns
the relationship between 'mood' and 'modality'.

Mood and modality

As used here, mood refers to a formally grammaticalized category of the verb

which has a modal function. Moods are expressed inflectionally, generally in
distinct sets of verbal paradigms, e.g. indicative, subjunctive, optative, im­
perative, conditional, etc., which vary from one language to another in respect
to number as well as to the semantic distinctions they mark. Modality, on the
other hand, is the semantic domain pertaining to elements of meaning that
languages express. It covers a broad range of semantic nuances—jussive,
desiderative, intentive, hypothetical, potential, obligative, dubitative, horta­
tory, exclamative, etc.—whose common denominator is the addition of a
supplement or overlay of meaning to the most neutral semantic value of the
proposition of an utterance, namely factual and declarative.1
In the terms of the framework set forth in Bybee and Dahl (1989),
modality is a semantic domain, while moods, as formal categories of gram­
mar, can be either cross-language gram types (e.g. conditional or subjunctive)
or language-specific categories (e.g. the Delayed Imperative in Buriat, an
Altaic language; note that we distinguish these two levels by use of an initial
upper case letter to signal language- or family-specific categories, retaining
lower case for cross-language gram types.)
Modality is expressed in language in a variety of ways: morphological,
lexical, syntactic, or via intonation. These are not mutually exclusive. Thus in
the Spanish sentence dudo que haya ganado el premio T doubt (that) he won
the prize', the 'dubitative' modality is conveyed redundantly by both the
lexical meaning of the main verb and the subjunctive mood of the subordi­
nate-clause verb. In this volume we will be concerned primarily with gram­
matical (morphological and syntactic) expressions of modality, including
forms that may be currently undergoing grammaticalization.
An Introductory Essay 3

Function and its relation to linguistic form

A second issue we wish to clarify at the outset of our discussion concerns the
relationship between domains of modality and their expression in natural
languages. In some approaches to modality, function is studied for its own
sake. In this volume we take the position that analysis of function should
explain distribution of form. That is, the modal categories we operate with do
not lead an autonomous existence in some abstract logical or semantic space;
rather, they correspond to—indeed are determined by—(a) the formal distinc­
tions made in particular languages, (b) documented pathways of language
change, and (c) prominent cross-language patterns of form-function correla­
tion.
In the area of modality, however, cross-language comparison has been a
difficult task, for several reasons. First, because the semantic/functional do­
main of modality is so broad; second, because modality, as we have discov­
ered, lends itself best to investigation in social, interactive contexts
(elaborated below); third, and conceivably most important, because of the
extent to which languages differ in their mapping of the relevant semantic
content onto linguistic form. A case in point that came up repeatedly in the
course of our discussions concerns the category irrealis and the nature of the
realis/irrealis distinction. The languages that came under our scrutiny differed
in terms of what they classify as realis and what they classify as irrealis,
prompting us to question whether the distinction is cross-linguistically valid
at all, and if so, whether it corresponds to a gram-type distinction, such as
perfective/imperfective, or whether it more closely resembles a supercategory
such as mood.
As noted above, it also became clear over the course of our discussions at
the symposium that many of the functions of modality are inextricably em­
bedded in contexts of social interaction and, consequently, cannot be de­
scribed adequately apart from their contextual moorings in interactive
discourse. Several papers in this volume explore particular modalities specifi­
cally as they occur in contexts of face-to-face communication. But even in
those that do not make explicit reference to contexts of social interaction, this
factor is nonetheless in evidence.
4 Modality in Grammar and Discourse

Modal categories and their associated nomenclature

Our linguistic understanding of modality has its roots in modal logic (a

branch of philosophy of language) and in particular in the distinction between
'deontic' and 'epistemic' modality. Modal logic has to do with the notions of
possibility and necessity, and its categories epistemic and deontic concern
themselves with these notions in two different domains. Epistemic modality
has to do with the possibility or necessity of the truth of propositions, and is
thus involved with knowledge and belief (Lyons 1977:793). Deontic modal­
ity, on the other hand, is concerned with the necessity or possibility of acts
performed by morally responsible agents (Lyons 1977: 823), and is thus
associated with the social functions of permission and obligation.
The epistemic notion is of considerable use to linguists, given that many
languages have grammatical markers which function explicitly to express an
evaluation of the truth of a proposition. Accordingly, most linguists under­
stand epistemic modality as expressing the degree of a speaker's commitment
to the truth of the proposition contained in an utterance. Thus, one way
epistemic possibility is expressed in English is by may and might (we may/
might lose the election), while epistemic necessity is expressed by must (they
must have won the election). However, as applied to natural language, there is
no reason to restrict the epistemic notion just to necessity and possibility, as is
traditional in philosophy of language. For one thing, commitment to the truth
of a proposition is often a matter of degree. For another, epistemic modality
can be seen as overlapping with, or even encompassing, another grammatical
category, namely evidentiality 2 For our purpose, the former expansion of the
epistemic notion will be adhered to (i.e. construal along a continuum) but not
the latter. The symposium from which this volume emerges deliberately
excluded the study of evidentials, which were the subject of an earlier meeting
in this same series (see Chafe and Nichols 1986).
Deontic modality has also proven to be a useful concept for linguists;
however, its translation into linguistic categories has not been as smooth as in
the case of epistemic modality (cf. Bybee 1985, Bybee, Pagliuca and Perkins
1991), As understood in philosophy of language, deontic modality focuses on
the notions of obligation and permission. It is found in directives that grant
permission (you may go now) or impose obligations (eat your vegetables!), as
well as in statements that report deontic conditions (Yeltsin should slow down
reforms in Russia; graduate students can check out books for the whole semester).
An Introductory Essay 5

A problem with the deontic notion for linguists, however, is the imperfect
nature of its fit with the corresponding linguistic categories that we encounter
in the world's languages as well as in language change: i.e., it is at once too
broad and too narrow. For one thing, unlike 'agent-oriented' modality—a
supercategory label that will be used by most papers in this volume in
preference to 'deontic' 3 —deontic modality fails to distinguish subcategories
that are expressed inflectionally (i.e. as grammaticalized moods), such as
imperative, from lexical ox periphrastic (i.e. auxiliary) expressions of obliga­
tion or permission. Furthermore, deontic modality as traditionally understood
excludes certain semantically related notions such as ability (physical and
mental) and desire that have linguistic expression similar to that of permission
and obligation. While one argument for the category 'deontic' might be the
well-documented pathway of change whereby deontic modals over time come
to acquire epistemic functions, in actual fact this change affects a broader
range of meanings than the term 'deontic' indicates.
The traditional division of modality into epistemic and deontic reveals
some interesting cases of polysemy in which the same form can be used for
both types of modality. Thus English may can express either deontic permis­
sion (you may come in now) or epistemic possibility (this may be your lucky
day!), while must can express deontic obligation (you must be here by seven)
as well as inferred probability (that must be the mailman at the door). A
diachronic view of this polysemy yields the observation, documented in many
languages, that so-called deontic meanings typically evolve into epistemic
meanings. 4
With regard to the difficulties attaching to 'deontic' as a supercategory
label, even this change from deontic to epistemic meaning in fact affects a
broader range of categories. While it is true that obligation markers may come
to be used for epistemic functions such as probability or inference, in the case
of permission markers it is not 'permission' per se that licenses a meaning of
epistemic possibility (e.g. in the case of may). Virtually all permission mark­
ers can be traced back to expressions of ability, permission being just one
sense of a more generalized 'root-possibility' meaning that arises from ability
(Bybee, Perkins and Pagliuca 1994). Root possibility predicates general ena­
bling conditions (e.g. it can take three hours to get there). These include
permission, which is a social enabling condition. Bybee (1988) has shown
that it is the root possibility sense that gives rise to epistemic possibility.
A second instance of categories other than narrowly-defined deontic
6 Modality in Grammar and Discourse

categories developing epistemic meanings is that of verbs indicating desire

(and obligation) evolving into futures. This development parallels the deontic
to epistemic shift in that a change occurs from a modal expression predicating
conditions on an agent—an 'agent-oriented' modality—to a modal expression
that has an entire proposition in its scope and communicates the speaker's
stance with regard to the truth of that proposition—a 'speaker-oriented'
modality.
For the reasons suggested above, Bybee (1985) proposed a change in the
categorial nomenclature of modals as follows: Agent-oriented modality
encompasses all modal meanings that predicate conditions on an agent with
regard to the completion of an action referred to by the main predicate, e.g.
obligation, desire, ability, permission and root possibility. Epistemic modal­
ity retains its traditional definition: epistemics are clausal-scope indicators of
a speaker's commitment to the truth of a proposition. Markers of directives,
such as imperatives, optatives or permissives, which represent speech acts
through which a speaker attempts to move an addressee to action, are called
speaker-oriented.
It will be observed that the distinction between agent-oriented and
speaker-oriented modalities cross-cuts the traditional category of deontic
modality. Agent-oriented modals include deontic statements (statements that
describe obligations and permission), while speaker-oriented modals include
speech-act types such as imperatives that impose conditions of obligation. A
prime motive for replacing the deontic category by an agent-oriented category
is that the latter better reflects general morphosyntactic trends in expression
type: i.e., there is a strong, quasi-universal tendency for agent-oriented modal­
ity to be expressed by verbs, auxiliaries or non-bound particles, whereas the
remaining two types (speaker-oriented and epistemic) are often expressed
inflectionally (Bybee 1985). A second universal pattern is diachronic, namely
the tendency referred to above whereby the agent-oriented modalities develop
predictably into the other two types (Bybee, Perkins and Pagliuca 1994).

Agent-oriented and epistemic modality

As noted above, many modal forms exhibit a systematic polysemy between

agent-oriented and epistemic meanings. The theoretical issue raised by such
polysemy is whether a form should be regarded as having a single underlying
basic meaning (a so-called invariant meaning) that combines with contextual
An Introductory Essay 7

factors to yield differing interpretations for individual tokens, or whether,

alternatively, it has a complex set of semantic properties that by definition
incorporate meanings derived from context. A more specific issue that this
polysemy raises is that of targeting the particular contextual factors that favor
one interpretation over the other. In this regard, Bernd Heine's paper focuses
on agent-oriented modals that can also express epistemic meanings, using
data from German. His analysis isolates contextual factors as well as concep­
tual properties resident in the meanings of the modals that operate together to
enable addressees to correctly choose an agent-oriented or an epistemic
reading. Jennifer Coates' contribution builds on Heine's analysis and dis­
cusses the particular case of root possibility (in English). Applying the
conceptual properties Heine proposes to contrast agent-oriented and epis­
temic interpretations of modals, Coates shows that this distinction is weaker
with respect to the modal domain of possibility than with respect to other
areas of modal meaning.
Analyzing the Spanish modals poder 'can' and deber 'ought to,' Car­
men Silva-Corvalán also argues for the importance of context in the interpre­
tation of modal meanings. However, she rejects the idea that these modals are
inherently polysemous, positing instead an invariant meaning for each one:
'does not preclude X' for poder and 'favors, requires or entails X' for deber.
These invariant meanings are more generalized than the agent-oriented mean­
ings and require supplementation from context to produce their actual inter­
pretations in particular utterances. Many of the contextual factors
Silva-Corvalán points to resemble those proposed by Heine.
Edith Bavin takes a diachronic and comparative approach to some of
these same issues, examining the development of obligation markers in West­
ern Nilotic languages. While obligation markers in these languages have
developed from sources different from those found in European languages
(i.e. from impersonal constructions rather than from agent-oriented modal
verbs), the same sorts of contextual factors influence addressees' interpreta­
tions of them as agent-oriented or epistemic.
Sherman and Phyllis Wilcox present one of the first analyses of modal­
ity in American Sign Language, showing once again the familiar pathway of
change whereby agent-oriented modals eventually acquire epistemic mean­
ings. The Wilcoxes also isolate a set of linguistic parameters that distinguish
the meanings of ASL modals, and describe further a set of gestural parameters
that iconically mirror the semantic properties of the modal markers.
8 Modality in Grammar and Discourse

The interactional basis of modality

In recent years, an increased understanding of many grammatical categories

has come about through examination of these categories in the actual contexts
in which they are used—what is referred to as 'discourse' or 'situation'
context. Whereas for the analysis of tense, aspect, transitivity, ergativity,
reference, etc., linguists have looked for the most part at narrative discourse,
and with good result, modality cannot be studied solely with respect to
narrative, since many modal functions surface only in face-to-face interactive
discourse. That is, they typically depend not just on a monologic speaker (the
narrator in narrative discourse), but on a dialogic (explicitly or by implication)
speaker-addressee interaction. This is the case in the imposing of obligations
(through statements or through directives), in the giving and receiving of
permission, and in exchanges of information, with appropriate expressions of
commitment to the truth of that information. In fact, modals can be viewed as
strategic linguistic tools for the construction of social reality, as demonstrated
by Julie Gerhardt (1985,1990) in detailed studies of the use of modals in child
language.
Two papers in this second section of the volume highlight the social
functions of modality by analyzing chronologically the development of the
functions to which children put modal forms. Soonja Choi's paper on sen­
tence-ending particles in Korean shows how children learn to integrate infor­
mation they have to offer into the aggregate of information shared by speaker
and addressee. Because of the strong interactive function of the Korean
particles and the way they figure in the construction of appropriate discourse,
these particles are acquired earlier by Korean children than are the more
purely epistemic markers of languages such as English. Jiansheng Guo's
study of the acquisition of Mandarin neng (roughly 'can') emphasizes that
children's command of a modality marker depends on their association of that
marker with particular contexts of interaction. The meanings of modals in
general, Guo argues, are rooted in the social, interactional functions of lan­
guage; and in the case of neng, which he interprets as functioning in various
ways as a challenge to the addressee, the information-exchanging function is
clearly subordinate to the interactive function.
The interactional functions of modals can also be observed in adult
language. Using texts that represent interactive discourse in written form (i.e.,
conversations in written texts of several different genres), John Myhill and
An Introductory Essay 9

Laura Ann Smith undertake to make cross-linguistic comparisons of the use

of obligation markers in languages of different types (English, Mandarin,
Biblical Hebrew and Hopi). An important finding of their study is that
obligation expressions can carry a range of subtle nuances that render them
appropriate for carrying out speech-acts other than simply imposing obliga­
tions or making reference to obligations already in force. These include
evaluating the effects of actions, explaining actions that might be construed as
impolite, expressing lack of sympathy, and persuading one's conversational
partner to do something.
Also included in this section is Frantisek Lichtenberk's comparative
study of the functions of the 'apprehensional epistemic' modality found in
various Austronesian languages. While the methodology of his study differs
from that of the three preceding papers in this section in not being based on a
data corpus of explicitly interactive discourse, the forms Lichtenberk investi­
gates—whose functions include issuing a warning to the addressee ('watch
out! you may get sick') as well as conveying the speaker's apprehension about
something that might happen to the addressee—are clearly suited to
negotiatory discourse. Moreover, the changes they undergo in certain of the
languages investigated clearly demonstrate their dependence on speaker-
addressee interaction.
We also include in this section a highly original—and unorthodox (in the
context of traditional analyses of modality)—paper by John Haiman that
helps delimit the range of grammaticalizable modalities by targeting a set of
speaker attitudes (the 'sarcastive', the 'guiltive', the 'mass-productive') that
as far as we know have never become grammaticalized as moods. Although
sarcasm functions like a mood, insofar as it expresses a speaker's attitude
toward the proposition of an utterance produced in a dialogic exchange, its
formal markers (segmental or suprasegmental) never seem to make it into the
service sector of natural languages, i.e. they never become grammatical.
Haiman speculates on why this is the case.

Irrealis modality and subjunctive

A term widely used in discussions of modality, especially with respect to

Native American and Indo-Pacific languages, is 'irrealis', often contrasted
with 'realis.' Irrealis refers to a very broad conceptual category that covers a
wide range of non-assertive modal meanings and receives formal expression
 10 Modality in Grammar and Discourse

in certain languages. Unfortunately, it is difficult to determine in many

instances whether the modal meaning of an utterance is contributed specifi­
cally by the so-called Irrealis marker or by some other element (lexical or
morphosyntactic) of the discourse context, in which case the Irrealis marker is
functionally redundant. This ambiguity calls into question the labeling of
certain language-specific forms as Irrealis markers. A similar problem arises
in the description of Subjunctives in European languages; like Irrealis mark­
ers, Subjunctives occur in a wide range of non-assertive contexts, and the
status of their semantic content is very much in question.
A second and not insignificant problem with irrealis as a cross-language
gram-type is the degree to which languages vary in their assignment of
notional categories to the grammatical category Irrealis. Each language that
operates with this category seems to make its own determination as to which
notional categories will be considered irrealis. Thus, for both irrealis and
subjunctive, it is difficult to circumscribe a focal meaning for the gram-type.
Several approaches to the latter problem are demonstrated in the papers of our
third section.
For Wallace Chafe, the realis/irrealis distinction is based on a funda­
mental assumption on the part of language users—an assumption which
presumably operates in the same way for users of any given language—that
some of their ideas belong to the domain of objective reality while others have
their source in the imagination. This basic cognitive principle of judged
reality vs. unreality is expressed formally by the Realis/irrealis distinction in
Caddo and the Northern Iroquoian languages, albeit in slightly differing
contexts and through the use of different formal devices. Chafe sees another
instantiation of this distinction in English speakers' judgments about the
referentiality or non-referentiality of indefinite arguments.
Marianne Mithun also treats Irrealis categories in Native American
languages in a paper that foregrounds the theoretical problem of the cross-
language variability of irrealis as a gram-type. Appealing to the same underly­
ing distinction as that described by Chafe, Mithun argues that this common
cognitive distinction gets applied in different ways in different languages, a
state of affairs which accounts for the fact that Irrealis functions vary so
widely across languages. She insists, however, that this cross-linguistic varia­
tion is not random, and explains the different language-specific construals of
irrealis described in her paper as the respective outcomes of differing dia-
chronic developments.
An Introductory Essay 11

The paper by Suzanne Romaine takes a diachronic look at the Tok Pisin
particle bai, which now functions primarily as a future marker but which also
has a variety of modal functions (as do most future markers). Drawing on an
extensive corpus of data (synchronic and diachronic, spoken as well as
written), Romaine traces the stages through which the clause-initial time
adverb baimbai 'by and by' grammaticalizes into the reduced pre-verbal
marker bai and in the process acquires a set of future and 'irrealis' functions.
Her study also points out the striking conformity of these developments to
cross-linguistically established trends.
Irrealis modality is also a primary focus of Suzanne Fleischman's paper,
summarized in the section below on the interaction of modality with other
categories of grammar.
As noted above, there are certain similarities between Irrealis categories
and the Subjunctives of European languages. Patricia Lunn's paper is repre­
sentative of recent work on the Spanish Subjunctive in its appeal to pragmatic
considerations—contextual and interactional factors—to account for the use
of Subjunctive vs. Indicative forms in several varieties of discourse. Lunn
shows that Subjunctive coding is not limited to unreal and non-assertive
propositions, as suggested in traditional accounts of the Spanish Subjunctive;
this mood can also be used to signal background information in literary texts,
and in journalistic discourse to mark particular information as 'common
knowledge'.

Modality and other categories of grammar

In the final section of this volume we group together four papers that in
different ways explore the interaction between modality and other domains of
grammar, specifically: negation, complementizers, past tense, and imperfec-
tive aspect. In these papers we see once again the crucial role that interactive
contexts play in shaping the meanings that result from the combinations of
grammatical categories.
Frank Palmer's contribution investigates the systematic irregularity
(this is not an oxymoron) that we find across languages in the behavior of
modals and in the meanings that emerge when modals appear under the scope
of negation. A particularly widespread irregularity involves the strategies
languages use to express the semantic notions of 'necessary-not' and 'not-
necessary'. This may involve use of a different verb altogether from the one
 12 Modality in Grammar and Discourse

normally used in affirmative contexts, as in English mustn't vs. needn't, or a

'displacement' of the negative marker for 'necessary-not' to a syntactic
position that corresponds formally to 'not-necessary', as in French and Italian
(Fr. il ne faut pas partir, It. non deve venire).
Zygmunt Frajzyngier's paper advances the claim that in various unre­
lated languages modality appears to be expressed by complementizers, which
function to code the modality of embedded clauses. More specifically, he
links the presence of one or more complementizers (treated as a parametric
variation within the GB framework), as well as their syntactic position in
embedded clauses, to the presence or absence (synchronic or diachronic) of
certain modalities in the sentence as a whole. Of particular interest in this
paper are cases of clauses with multiple complementizers expressing different
types of modality, e.g. agent-oriented and epistemic. The paper also has
implications for formal syntactic theories that posit COMP as a component of
the sentence.
When agent-oriented modals combine with past tense the resulting unit
often undergoes a meaning change, losing the past-tense component of its
meaning and coming to signal a weakened version of its original modal
meaning in the present tense. Joan Bybee's paper explores the reasons for this
development by studying the uses of would and should in texts from Middle
English and Early Modern English. Bybee argues that these modals lose their
past sense because of the implication that a modal condition in past tense
continues into present time, pointing to the conclusion that modal meaning is
heavily influenced by the interactive contexts in which it is used.
The last paper in the volume, by Suzanne Fleischman, surveys a wide
range of evidence pointing to an overlap (synchronic and diachronic) between
the aspectual gram-type imperfective and irrealis modality, and poses the
question of why, in so many unrelated languages, verb forms marked for
imperfective aspect come to acquire meanings subsumable under the broad
modal heading of irrealis. Since none of the languages Fleischman refers to
has a formally grammaticalized Irrealis category, and in light of the notorious
elusiveness of irrealis pointed out above, her definition of irrealis is worth
noting here. She characterizes 'irrealis' as prototype category, at the semantic
level, expressing a spectrum of meanings that signal a speaker's lack of belief
in or lack of commitment to any of the following: the reality or referentiality
of a situation; the possibility that an agent's wishes, hopes or intentions will
effectively be realized; the authenticity of an utterance or a chunk of dis-
An Introductory Essay 13

course; or the normalcy of a discourse or of a communicative situation. These

parameters are illustrated with data from various languages and discourse
genres.
So, here goes, readers. We hope you enjoy the papers. And don't be put off by
the realization that mood is a grammatical category with an attitude! We'll
get it straightened out one of these days.

NOTES
* We are grateful to Greg Thomson for assisting the editors and authors in manuscript
preparation and copy editing. The indexes were prepared by Lisa Dasinger and Jacki
Trademan.
1 In this volume we avoid the term 'mode' because of the problematic ambiguities it
presents, being used with widely different meanings in the grammars of different
languages. In many European languages it translates what is here referred to as 'mood',
while in the grammars of certain non-European languages it is used to label categories
whose meanings fall under the headings of tense and aspect. We are aware, of course,
that languages often bundle tense, aspect, and mood information into portmanteau
morphology, thereby making it difficult to decide how to label such categories.
2 The term 'evidential' was first introduced by Jakobson (1957) as a tentative label for a
verbal category that indicates the source of the information on which a speaker's
statement is based. As currently understood, evidentiality covers a range of distinctions
involved in the identification of the source of one's knowledge. Various languages have
grammaticalized evidential markers indicating whether or not the speaker vouches
personally for the information contained in a statement. (See Chafe and Nichols 1986,
Willett 1988).
3 In place of the traditional distinction of linguistically-relevant modals into 'epistemic'
and 'deontic', Bybee (1985) recategorizes the modals into 'agent-oriented,' 'speaker-
oriented', and 'epistemic'. These categories will be defined and elaborated on below.
Other categories distinguished in modal logic, e.g. 'dynamic' and 'alethic' modalities
(cf. Lyons 1977:791, Palmer 1986:102-103), will not be discussed here, being less
germane to the analysis of modality in natural language.
4 The earlier 'deontic' meanings may or may not be preserved. English must, for example,
retains its obligative meaning, whereas might has lost its earlier abilitative meaning.

REFERENCES
Bybee, Joan L. 1985. Morphology: A Study of the Relation between Meaning and Form
(=Typological Studies in Language, 9) Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan L. 1988. "Semantic Substance vs. Contrast in the Development of Grammati­
cal Meaning." Proceedings of the Berkeley Linguistic Society. 14.247-264.
Bybee, Joan & Östen Dahl. 1989. "The Creation of Tense and Aspect Systems in the
Languages of the World". Studies in Language 13.51-103
14 Modality in Grammar and Discourse

Bybee, Joan L., William Pagliuca, & Revere D. Perkins. 1991. "Back to the Future".
Approaches to Grammaticalization. ed. by E. C. Traugott &. B. Heine, vol. 2, 17-58.
Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan L., Revere D. Perkins & William Pagliuca. 1994. The Evolution of Grammar:
Tense, Aspect and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Chafe, Wallace L., & Johanna Nichols. 1986. Evidentiality: The Coding of Epistemology
in Language. Norwood, N.J.: Ablex.
Gerhardt, Julie. 1985. "On the Use of will and gonna: Toward a Description of Activity
Types for Child Language". Discourse Processes 8.143-75.
Gerhardt, Julie. 1990. "The Relation of Language to Context in Children's Speech: The
Role of hafta Statements in Structuring 3-Year-Old's Discourse". IPrA Papers in
Pragmatics A. 1-57.
Hopper, Paul, ed. 1982. Tense-Aspect: Between Semantics and Pragmatics. Amsterdam:
John Benjamins.
Jakobson, Roman. 1957. Shifters, Verbal Categories, and the Russian Verb. Cambridge,
Mass: Russian Language Project, Department of Slavic Languages and Literatures,
Harvard University. (Repr. The Selected Writings of Roman Jakobson. Vol 2.130-147
[1971] The Hague: Mouton.)
Lyons, John. 1977. Semantics , vol. 2. Cambridge: Cambridge University Press.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Willett, T. 1988. "A Cross-Linguistic Survey of the Grammaticization of Evidentiality".
Studies in Language. 12.51-97.
 I
Agent-Oriented and Epistemic Modality
 Agent-Oriented vs. Epistemic Modality
Some Observations on German Modals1

Bernd Heine
University of Cologne

1. Introduction

Recent research on the domain of modality has focussed in particular on the

nature of one basic distinction, that between what has variously been referred
to as deontic, root, objective, pragmatic or agent-oriented modality on the one
hand and subjective, hypothetical, or epistemic modality on the other.2 A
number of generalizations have been proposed about the nature of agent-
oriented as opposed to epistemic modality, such as the following:
a. In many languages, agent-oriented and epistemic meanings are
expressed by means of one and the same expression.
b. Wherever there is adequate historical evidence available on the
evolution of a language, agent-oriented meanings have been shown
to be older than epistemic ones (Shepherd 1982, Bybee and
Pagliuca 1985, Traugott 1989).
c. Epistemic meanings are said to be more strongly subjective than
agent-oriented ones, and the transition from agent-oriented to epis­
temic uses of a given linguistic expression has been described as
involving subjectification (cf. Langacker 1985), whereby mean­
ings tend to become increasingly based in the speaker's subjective
belief state/attitude toward the proposition (Traugott 1989:35).
d. In the ontogenesis of language, agent-oriented modality develops
prior to epistemic modality (but see Soonja Choi, this volume for
different observations). For example, for both English and German
 18 Bernd Heine

children, modal verbs at first serve agent-oriented rather than epis-

temic functions (Stephany 1989:4).
e. Similarly, in second language acquisition, agent-oriented uses of
modal elements tend to appear earlier than epistemic ones (Ramat
1992).
There are a number of questions that the study of these two kinds of
modality raises, such as the following:
(i) What factors can be held responsible for the fact that linguistic
expressions for agent-oriented modality are extended to also ex­
press epistemic modality?
(ii) If a linguistic item is used for the expression of both agent-oriented
and epistemic modality, are we dealing with an instance of ho-
monymy or of polysemy, that is, does that item simultaneously
belong to two different categories or only to one? The former view
is favored by Coates (1983), Palmer (1986), as well as a number of
other authors, while the latter is argued for by Sweetser (1982),
Traugott (1989) and others.
(iii) Is the transition from one kind of modality to another discrete/
discontinuous or gradual/continuous?
(iv) How can the shift from agent-oriented to epistemic modality be
described in a model based on grammaticalization theory?
While these questions have been addressed in a good deal of research
over the past decade, none has been answered satisfactorily, nor will the
present paper attempt to provide conclusive answers. Most discussions relat­
ing to these questions have centered around modality in English. This paper
will focus on a language different from but closely related to English, and its
aim is to provide some fresh data that might be relevant to answering the
above questions.

2. German modals

2.1 German and English modals compared

A number of authors have pointed out that German modals differ from their
English counterparts essentially in being more verb-like (Abraham
Agent-Oriented vs. Epistemic Modality 19

1989:357); some (cf. Jenkins 1972:9-12; Steele et al. 1981; Abraham 1992)
go so far as to argue that a categorial boundary separates German and English
modals, the latter belonging to a category called AUX and the former to that
of Verb. German modals such as können 'can' or müssen 'must', etc. differ
from their English counterparts, e.g., in the following ways:

a. They exhibit inflections (e.g., ich kann T can', du kannst 'you

can'). 3
b. They may appear in non-finite constructions (e.g., um kommen zu
können 'in order to be able to come').
c. They may iterate (e.g., er will arbeiten können 'he claims that he
can work').
d. They are not restricted to any fixed order with respect to each other
or to the perfect auxiliary haben 'have' (e.g., er hat arbeiten
gekonnt 'he has been able to work' vs. er kann gearbeitet haben 'he
may have worked').
e. They may also occur in certain constructions without any other
verb in the same clause (e.g., er kann Handstand 'he can do a
handstand').
f. "Auxiliary reduction" does not apply to them.4
There are, however, a number of properties that German modals share
with their English counterparts, e.g. (cf. Helbig and Buscha 1988:107ff.; 123ff.):
(i) They express a restricted range of grammatical functions.
(ii) They form a closed set of entities.
(iii) They normally do not passivize.
(iv) They do not form imperatives.
(v) They typically require main verbs, rather than noun phrases or
adverbial phrases, as complements.
(vi) In their presence, main verbs are used in a non-finite (infinitival)
form.
(vii) They lack a present tense 3 sg. inflection (-s in English and -t in
German).
(viii) They do not take the infinitive marker (zu) to introduce the main
verb.
(ix) Finally, and most importantly in this context, German and English
modals are both used to express agent-oriented as well as epistemic
meanings.
20 Bernd Heine

What this suggests is that we are dealing with a difference in degree

rather than in kind: while German modals are more verb-like than their
English cognates, the two are similar enough to be comparable. For more
details, see Heine (1993).

2.2. A survey

In the present section, eight German modals are looked at with a view to
determining how and to what extent they are associated with agent-oriented/
deontic and epistemic modality, respectively, or, to use the terminology of
German grammarians, with "objective" and "subjective" modality. The mo­
dals in question are:
müssen 'must'
sollen 'should'
können 'can'
wollen 'want, will'
mögen 'like, may'
dürfen 'be allowed to, may'
möchten 'want to, would like to'
werden 'will'
One may wonder what justification there is to include werden, essentially
a future tense marker, within the paradigm of modal auxiliaries. While I do
not wish to argue that werden is a modal, as has been done elsewhere (see
especially Vater 1970), the main reason for including it here is that like "true
modals" it exhibits a distinction between non-epistemic and epistemic uses.
Note furthermore that können, like English can, has two non-epistemic
senses: root possibility and permission (cf. Bybee, Perkins, and Pagliuca
1994). These two senses will not be distinguished in the quantitative analysis
that follows; our concern will be essentially with the distinction between non-
epistemic and epistemic modality, rather than between agent-oriented and
epistemic modality. These distinctions will be clarified below.
In the association of the eight modals with these two categories of
modality in given utterances, the following three situations will be distin­
guished, where "A" stands for agent-oriented and "E" for epistemic modality:
Agent-Oriented vs. Epistemic Modality 21

A, E: The relevant utterance is automatically associated with either an A

or an E interpretation, with no further contextual clues required,
i.e., either A or E constitutes the "basic meaning" or focal sense
(see Heine, Claudi and Hünnemeyer 1991) of the utterance con­
cerned.
A l , El : The relevant utterance does not have a focal sense of A or E, but is
associated with specific linguistic or extra-linguistic contexts that
are suggestive of either an A or an E interpretation. We may say
that in such cases A or E constitutes a non-focal sense.
A2, E2: The relevant utterance is not automatically associated with either A
or E. There are, however, highly specific contexts where an A or an
E interpretation is conceivable. In such cases, A or E will be said to
constitute a marginal sense.
This threefold classification is somewhat crude; a more fine-grained
typology would be desirable. For the present analysis, however, this classifi­
cation is sufficient to study the problems at issue here. I illustrate the three
situations with two examples.5
(1) Er muß mindestens 1,80 m sein.
he must [Link] 1.80 metres be
'He must be at least six foot.'
a. [E: "On the basis of the evidence available I am led to conclude
that..."]
b. [A: "They are looking for a new goal-keeper (he has to be at
least six foot)"]
The most likely reading of muß in (1) is the epistemic one (cf. (la), hence
(1) is said to have a focal epistemic sense (E). (1) has, however, a non-focal
agent-oriented sense (Al) in addition, which is foregrounded, e.g., in a
context like (lb).
An example of a marginal sense is provided by the utterance in (2) below.
While (2) has a focal A sense ("He is obliged to come"), contexts can be
conceived of that would invite an E2 interpretation. A context like (2a), for
example, calls up the marginal epistemic sense, conveying the meaning:
"I have every reason to conclude that he is coming".
22 Bernd Heine

(2) Er muß kommen.

he must come
'He has to cóme.'
a. ["I can already hear his voice (so he must be coming)"]

In what follows, quantitative data are presented that bear on the question
of whether and to what extent certain standardized utterances are associated
with the two kinds of modality. To this end, index values are calculated on the
basis of the three-fold classification proposed above: a value of 1.00 is given
to an utterance whenever a focal sense is involved, 0.66 in the case of a non-
focal sense, and 0.33 in the case of a marginal sense. I illustrate this procedure
with reference to examples (1) and (2) above: the utterance in (1) (Er muß
mindestens 1.80 m sein) would receive an index value of 1.00 for its focal
epistemic sense and of 0.66 for its non-focal agent-oriented sense. Utterance
(2), on the other hand, would have an index value of 1.00 for its focal agent-
oriented sense and of 0.33 for its marginal epistemic sense. Values are
calculated for declarative (e.g., Er muß kommen 'He must come'), interroga­
tive (Muß er kommen? 'Does he need to come?'), and negative utterances (Er
muß nicht kommen 'He need not come'), for main verb perfect forms (Er muß
gekommen sein 'He must have come'), auxiliary past tense forms (Er mußte
kommen 'He had to come'), as well as for first- (Ich muß kommen 'I must
come'), second- (Du mußt kommen 'You must come'), and third-person
subject utterances (Er muß kommen 'He must come'). The sum totals are
divided by the total number of utterances considered, thereby establishing
average index values for agent-oriented and epistemic modalities. The maxi­
mum value attainable is 1.00, the minimum value 0 (zero). For example, a
modal will receive the maximum value of agent-oriented modality if in all
utterances considered it has a focal agent-oriented sense, and the minimum
value if it has neither focal, non-focal or marginal agent-oriented senses; see
APPENDIX for details. 6
A number of calculations have been carried out on the basis of selected
parameters; the results obtained are summarized in Tables 1 through 4. In
Table 1 the eight modals are compared in terms of the degree to which they
correlate with each of the two kinds of modality. As the index values suggest,
the modals can be arranged along a scale for each of the two modalities. At
Agent-Oriented vs. Epistemic Modality 23

Table 1. Index of modality of eight German modals (Affirmative/negative, declarative/

interrogative, present/past tense uses; see APPENDIX)
Index value of modality
Agent-oriented Epistemic
Mochten 0.92 0
Dürfen 0.74 0
Wollen 0.75 0.15
Müssen 0.51 0.37
Mögen 0.47 0.37
Werden 0.59 0.46
Können 0.46 0.50
Sollen 0.44 0.50

one end of the scale are möchten and dürfen, which have the highest values for
agent-oriented modality and exhibit no epistemic uses, at least not in the
corpus considered. [FN 7] At the other end we find können and sollen, which
have the lowest values for agent-oriented and the highest values for epistemic
modality.
The extent to which these modals are associated with agent-oriented and
epistemic modality appears to correlate with their relative degree of grammat-
icalization in general and with prototypical 'verbiness' in particular: thus, the
highest A values and the lowest E values are found with the items most
prototypically verb-like. Wollen 'will', for instance, is more verbal (has more
verbal properties) than modals like müssen 'must', können 'can', or sollen
'should', in that it can still take subordinate clauses introduced by daß 'that'
as complements and, accordingly, does not require subject identity between
the auxiliary and the main verb, as in (3). Also it can be passivized, as in (4)
(Vater 1970).
(3) Hans will, daß Anna zu Hause bleibt.
Hans wants that Anna at home stays
'Hans wants Anna to stay home.'
(4) Von Hans wird gewollt, daß Anna zu Hause bleibt.
by Hans is wanted that Anna at home stays
'Hans is requested to leave Anna at home.'
24 Bernd Heine

Table 2. Index of modality of four German expressions (on the basis of eight German
modals; see Appendix)
Index value of modality
Agent-oriented Epistemic
a. Sie kommt. 'She is coming.' 0.87 0.02
b. Sie weiß es. 'She knows it.' 0.77 0.32
c. Sie hat viel Geld. 'She has a lot of money.' 0.67 0.27
d. Sie kennt ihn. 'She knows him.' 0.37 0.59

Table 3. Index of modality of eight German modals according to various grammatical

distinctions (see Appendix)
Type of proposition Index value of modality
Agent-oriented Epistemic
a. Perfect
(e.g., Er hat kommen müssen. 'He had to come.') 0.75 0
b. Interrogative
(e.g., Muß er kommen? 'Does he need to come?') 0.72 0.11
c. Negative
(e.g., Er muß nicht kommen. 'He need not come.') 0.74 0.26
d. Past
(e.g., Er mußte kommen. 'He had to come.') 0.62 0.20
e. Present
(e.g., Er muß kommen. 'He must come.') 0.60 0.43
f. Main verb perfect
(e.g., Er muß gekommen sein. 'He must have come.') 0.19 0.55
g. Main verb progressive
(e.g., Er muß am Kommen sein. 'He must be coming.') 0.02 0.64

Table 4. Index of modality of eight German modals according to grammatical person of

the sentence subject (Affirmative/negative, declarative/interrogative, present/
past uses; see Appendix)
Index value of modality
Agent-oriented Epistemic
1st person subject 0.70 0.18
2nd person subject 0.69 0.25
3rd person subject 0.57 0.35
Agent-Oriented vs. Epistemic Modality 25

The behavior of the modals also differs greatly in accordance with the
type of main verb figuring in the utterances. Thus, verbs of action and telic
verbs associate predominantly with agent-oriented modality; cf. utterance (a)
in Table 2, involving the motion verb kommen 'come'. The situation is more
complicated in the case of stative verbs: while some, like wissen 'know
(information)' and haben 'have', gravitate toward agent-oriented uses,
kennen 'know, be acquainted with', has a majority of epistemic uses.
The modal behavior of the eight items considered also varies in accord­
ance with the type of proposition and the tense and aspect in which the modals
occur, as can be seen in Table 3. According to this table, in Perfect construc­
tions the modals are confined to agent-oriented uses (a). Even with verbs such
as kennen, which generally associate strongly with epistemic modality, an
epistemic reading is unlikely once the modal auxiliary has a Perfect form.
Furthermore, in Interrogative (b) and Negative (c) propositions and in the
Past tense (d) the modals correlate primarily with agent-oriented rather than
with epistemic modality. Epistemic uses prevail, however, if the main verb is
in the Perfect (f), and Progressive constructions8 are almost entirely associ­
ated with epistemic modality (g).
Finally, the interpretation of an utterance is also affected by the gram­
matical person of the subject. As Table 4 suggests, an utterance is most likely
to have an agent-oriented interpretation when the subject referent is first-
person and least likely when it is third-person; conversely, epistemic modality
correlates most strongly with third-person and least strongly with first-person
subjects.

3. Context

3.1 The linguistic context

The quantitative data on German modals presented in the preceding section

allow us to isolate several linguistic factors that appear to be relevant for
characterizing the uses of the two kinds of modality. These data suggest that
an utterance is very likely to have a focal agent-oriented sense if:
a. any of the modals möchten, dürfen or wollen is involved;
b. the main verb is an action or a terminative verb;
c. the modal is in the perfect or past tense.
26 Bernd Heine

Furthermore, a modal is more likely to have agent-oriented than epistemic

modality as its focal sense if:
d. it occurs in interrogative rather than in declarative utterances;
e. it occurs in negative rather than in affirmative utterances;
f. the subject is first or second (rather than a third) person.
Thus, the more of the linguistic properties (a) through (f) an utterance
has, the more likely it is to express agent-oriented rather than epistemic
modality. Conversely, a focal epistemic sense is more likely if the main verb
is in a progressive or perfect form or is stative.
What these observations suggest is that the agent-oriented vs. epistemic
distinction works in roughly the same way with German as with English
modals. For example, according to the findings on English modals presented
by Coates, the following associations can be observed: (a) All agent-oriented
modals can occur in interrogative constructions where epistemic modals are
not normally found (1983:244-245). (b) Epistemic but not agent-oriented uses
are associated with the progressive aspect (1983:246). (c) The core meanings
of agent-oriented modals are crucially associated with agentivity, that is, the
action referred to in the main predication is carried out by someone (or
something) using their own energy (Coates 1983:245). (d) Epistemic meaning
is typically associated with existential subjects and stative verbs (1983:245;
see also Bybee, Perkins, and Pagliuca 1994). We will return to these cross-
linguistic observations below.

3.2 Contextual frames

While the above linguistic characterization allows us to predict within limits

whether a given utterance will receive an agent-oriented or an epistemic
interpretation, it does not account for a considerable number of actual uses of
German modals. Many times, when queried whether a given utterance was
suggestive of agent-oriented or epistemic modality, an informant's reaction
was to search for a relevant context based on past experience. If the informant
could remember a context that matched the utterance in question, then he or
she would answer the question with reference to that experience. If not, then
he or she would be likely to declare the utterance unacceptable. Thus, rather
than semantic, syntactic or other rule-governed behavior, it was the presence
or absence of a cluster of related experiences, let us call it the contextual
Agent-Oriented vs. Epistemic Modality 27

frame, that seemed to be crucial in deciding between the two modalities.

Consider the utterance in (5):
(5) Das Bier sollte kalt sein.
the beer should cold be
'The beer should be cold.'

In accordance with the observations made in 3.1, (5) is strongly suggestive of

a focal epistemic sense since there is an inanimate subject and a stative main
verb. Nevertheless, the majority of our informants, all native speakers of High
German, described (5) as having a focal agent-oriented sense. What appears to
account for the agent-oriented interpretation of (5) is its association with a
contextual frame like a party situation where someone is offered a warm beer.
One possible reaction would be to answer with (5), and in such a case, (5) can
only receive an agent-oriented interpretation. In accordance with the terminol­
ogy introduced in Section 2, the contextual information just presented may be
said to constitute the focal contextual frame for (5). A non-focal frame would
be provided, e.g., by (5a) which triggers an epistemic interpretation of (5).
5a ["I put the drinks into the fridge more than two hours ago
(so the beer should be cold by now)"]

Coates (1983:233) notes that the interpretation of English modals as

agent-oriented (that is, "root" in her terminology) depends in most cases on
the presence of agentivity. This also applies to a large extent to German
modals, though agentivity need not be, and frequently is not, explicitly
expressed; what matters is the contextual frame associated with the use of the
modal. The utterance in (5) has no formally expressed agent; its association
with focal agent-oriented modality derives from contextually recoverable
agentivity, i.e., the speaker wishes to urge someone to put the beer into the
fridge, or to replace the bottle of warm beer by a bottle of cold beer. What this
suggests is that any attempt at characterizing the relationship between the two
types of modality strictly in terms of the linguistic categories that typically
correlate with each one is likely to achieve little more than probabilistic
approximations; it is unlikely to explain the relationship.
The notion of a contextual frame is similar to what others have called an
'inferential schema', or simply a 'frame', i.e. a body of knowledge evoked by
the language user in order to provide an inferential basis for the understanding
28 Bernd Heine

of an utterance (cf. Levinson 1983:281). It is beyond the scope of the present

paper to define the parameters that are relevant for classifying contextual
frames as either focal or non-focal; presumably they have to do with factors
such as the following:
a. Contextual clues: Focal frames require a minimum of contextual
clues, they are triggered spontaneously without involving major
mental effort or imagination. Non-focal frames, on the other hand,
require a certain amount of mental energy.
b. Knowledge of the world: An example like (6) has a focal epistemic
sense, since we cannot manipulate time in such a way as to license
an agent-oriented interpretation.
(6) Es muß drei Uhr sein.
it must three o'clock be
'It must be three o'clock.'
c. Social norms: We believe in a positive world and expect other
people to behave in a socially acceptable way. Therefore, while
(7a) may receive either an agent-oriented or an epistemic interpre­
tation, (7b) receives only an epistemic sense, since we do not
expect someone to want someone else to behave in a socially
disapproved way.
(7) a Er muß tapfer sein.
he must brave be
(i) 'He has to be brave.'
(ii) 'He must be brave.'
b Er muß feige sein.
he must cowardly be
'He must be a coward.'
Conversely, utterances having human agents as subjects and action verbs
as predicates, that is, structures typically associated with agent-oriented mo­
dality, trigger a focal epistemic meaning if they are associated with contextual
frames that discourage an agent-oriented interpretation, as in (8).
(8) Er muß seine Frau schlagen.
he must his wife beat
'He is said to beat his wife.'
Agent-Oriented vs. Epistemic Modality 29

These are but a few of the factors that may be relevant to the evocation of
contextual frames, possible additional factors being frequency and/or inten­
sity of experience, perceptual salience, propensity for memory storage, or the
stereotypes an utterance tends to evoke.

3.3 Conceptual properties

In addition to contextual frames, the choice between the two kinds of modality
depends crucially on certain conceptual properties. Agent-oriented uses of
German modals are associated with the following properties:
a. There is some force (F) that is characterized by an "element of
will" (Jespersen 1924:320-1), i.e., that has an interest in an event
either occurring or not occurring.9
b. The event is to be performed typically by a controlling agent (C).
c. The event is dynamic (D), i.e., it involves the manipulation of a
situation and is conceived of typically as leading to a change of
state.
d. The event has not yet taken place at reference time, i.e., its occur­
rence, if it does in fact take place (see (e) below), will be later than
the reference time (L).
e. The event is non-factual (Palmer 1986:96), though there is a certain
degree of probability that it will occur (P).
Take utterance (9), for example, where F is the speaker (9a) or some
other force the speaker has in mind (9b), and C is the sentence subject er. The
event is dynamic (D), as can be concluded from the use of the action verb
kommen, and is supposed to take place later than at reference time (L).
Finally, the modal muß expresses a high degree of probability (= strong
obligation) that the event will in fact take place (P).
(9) Er muß kommen.
he must come
'He has to come.'
a. [A: "I insist that he comes"]
b. [A: "His boss insists that he comes"]
Differences between the various modals exist with regard to the degree of
probability that the event will take place: the probability is low in the case of
30 Bernd Heine

können and dürfen (permission) but high in the case of müssen (strong
obligation), with the remaining modals being intermediate. But the modals
differ also in other respects. There are essentially three kinds of relevant
participants: the speaker (S), the modal force (F), and the agent (C), the last of
these being typically, though not necessarily, coded as the sentence subject.
While F is different from C in the case of modals such as müssen 'must',
sollen 'shall, should' or können 'can' (= "subject-external modality"), as can
be seen, e.g., in (9), it is identical with C in the case of the modals wollen,
mögen, and möchten (= "subject-internal modality"), as in (10), where the
subject ich 'I' is both the modal force F and the agent C, and in this instance
also S.

(10) Ich möchte kommen.

I want, to come
'I would like to come.'
F can refer to either the speaker or to some other human being, or even to
a non-human entity. Thus, in (11), F may be either a human authority, e.g., the
speaker, or a non-human (e.g., religious, institutional, or moral) power.
(11) Du darfst nicht stehlen.
you may not steal
'Thou shalt not steal.'
There is no necessary relationship between F and C on the one hand, and
certain kinds of nominal referents on the other. For example, in the agent-
oriented reading of (5), repeated below for convenience, it remains unclear, in
the absence of additional contextual clues, whether F refers to the speaker or
to some other force, or whether C refers to the hearer, to another agent, or to
no explicit agent at all.
(5) Das Bier sollte kalt sein.
the beer should cold be
'The beer should be cold.'
While there are no compelling clues as to the referential identity of F in
(5), nevertheless, some kind of F is implied, as can be derived, e.g., from the
fact that a possible response to (5) would be to inquire about the referent of F
by asking: Wér möchte, daß das Bier kalt ist? 'Who wants the beer to be cold?'
Agent-Oriented vs. Epistemic Modality 31

While prototypical instances of agent-oriented modality are character­

ized by the presence of the properties listed above, prototypical instances of
epistemic modality lack all properties except P. Thus, in the epistemic reading
of (5), there is neither F nor C, no D is involved, and the later-than-reference-
time constraint (L) does not apply (see (14) below).
Since P is shared by the two kinds of modality, it will not be considered in the
remainder of this paper.

3.4 On the transition from agent-oriented to epistemic meanings

While we can distinguish the two kinds of modality on the basis of the
conceptual properties C, F, D and L, these properties are not of equal impor­
tance. Consider, for example, utterance (12b). The use of muß does not entail
L; that is, there is no later-than-reference-time constraint involved: the action
of schlafen can be assumed to take place over an interval that includes
reference time, yet (12b) has a focal agent-oriented sense.
(12) a. Warum ist Klaus nicht hier?
why is Klaus not here
'Why is Klaus not here?'
b. Er muß schlafen.
he must sleep
'He has to sleep.'
It might seem as if F is also irrelevant in (12b) since no explicit force is
mentioned. Yet one could inquire about a possible F by asking: Wér ist dafür
verantwortlich, daß er schlafen muß? 'Who is responsible for the fact that he
has to sleep?' Thus, F may be said to be potentially present in (12b), albeit
backgrounded.
In examples like (12b), where L does not apply, the difference between
agent-oriented and epistemic modality becomes minimal: While (12b) has a
focal sense of agent-oriented modality, it may also be understood to convey an
epistemic meaning, i.e. 'he must be sleeping.' In such cases we are dealing
with what Coates refers to as merger: the agent-oriented and the epistemic
senses of (12b) are in a both/and relationship, they are mutually compatible;
in order to understand (12b) it is therefore not necessary to decide which of the
two senses is intended since they are not mutually exclusive (see Section 4).
32 Bernd Heine

Furthermore, there are contexts in the use of müssen 'must' where both C
and L are absent and D is largely irrelevant, as in (13). Verbs that behave in a
similar way to frieren 'freeze, be cold' are leiden 'suffer', weinen 'cry', or
gähnen 'yawn'.
(13) Er muß frieren.
he must freeze
'He is freezing/feeling cold.'
In examples like (13), F is some unidentified force, and we are dealing
with an instance of focal agent-oriented and non-focal epistemic modality.
Once F is eliminated, however, an agent-oriented interpretation is ruled out:
no F is implied in (14), which therefore expresses exclusively epistemic
modality.
(14) Ihn muß frieren.
him must freeze
'He must be freezing/cold.'
What the observations made in this section suggest is that of the four
properties considered, only F is obligatorily connected with agent-oriented
modality: modal utterances that lack F lack the "element of will" to which
Jespersen (1924:320-21) referred. In such cases we are dealing with an
epistemic rather than an agent-oriented sense.
Typically, the lack of F will leave all other components of the modal
concept unaffected. Thus, P remains the same whether or not F is present: the
degree of probability that the situation described in the utterance will obtain is
high in the case of müssen 'must' and low in the case of können 'can',
irrespective of F. Take, for example, the modal sollen 'shall, should': in the
agent-oriented interpretation of (15), F may refer either to the speaker (15a)
or to another person (15b), while in the epistemic interpretation of (15), the
speaker is excluded as source of the modal force (15c); the modal source is
associated with someone other than the speaker, as suggested by the fact that
(15d) would not be an appropriate paraphrase of (15).

(15) Sie soll heute zu Hause sein.

she should today at home be
'She should be at home today.'
Agent-Oriented vs. Epistemic Modality 33

a. [A: "I want her to be at home today"]

b. [A: "Someone wants her to be at home today"]
c. [E: "Someone claims that she is at home today"]
d. [ *"I claim that she is at home today"]
In many of its agent-oriented uses, sollen excludes the speaker as a
source of F, as can be seen in (16), where the subject is a first-person referent.
(16) Ich soll kommen.
I should come
'I should come.'
[A: "Someone wants me to come"; *"I want to come"]

3.5 Discussion

The distinction between the two categories of modality has been associated
with a number of factors. First, as pointed out above, one major distinguishing
property is that agent-oriented senses are likely to refer to situations that are
supposed to obtain later than at reference time (= property L). Thus, an
utterance like (17) (= (5)) has future reference when used in its agent-oriented
sense (17a) but present reference when used epistemically (17b).
(17) Das Bier sollte kalt sein.
the beer should cold be
T h e beer should be cold.'
a [A: "I want the beer to be cold (so you'd better put it into the
fridge again")]
b [E: "I have reason to assume that the beer (standing in front of
me) is cold"]
Similar observations have been made for English modals. Coates
(1983:235), for example, notes that, with the exception of root can, English
modals in their agent-oriented uses always have future reference (see also
Palmer 1986:97; Bybee, Perkins, and Pagliuca 1994):
[...] one of the conditions of using Root MUST, SHOULD or OUGHT (=
'Obligation') or Root MAY or CAN (= 'Permission') is that the speaker
believes that the action referred to in the main predication has not already
been achieved. In other words, commands, recommendations and permis­
sion-granting utterances all refer to an action which will be carried out at a
time subsequent to the utterance (Coates 1983:233).
34 Bernd Heine

As noted in 3.4, the association of agent-oriented modality with deictic time is

not as rigid in the case of German modals as Coates suggests it is for English
modals. Note further that there are contexts where the later-than-reference-
time constraint does not apply, e.g., in the case of modals used in the past
tense, as in (18), where exclusively agent-oriented modality is involved.
(18) a Er mußte es tun.
he [Link] it do
'He had to do it.'
b Er wollte zu Hause sein.
he wanted at home be
'He had intended to be at home.'
What this suggests is that the later-than-reference-time constraint is a
concomitant, rather than a defining, property of agent-oriented modality.
Furthermore, Bybee, Perkins and Pagliuca (1994) observe that the two
modalities expressed by English must correlate with mutually exclusive lin­
guistic environments: in the Past tense, in the Present tense with a stative verb,
and in the Progressive, must can only have an epistemic ("inferred certainty")
reading, while in the non-past with dynamic verbs, must is only agent-oriented
and refers to the future. Apart from the strong correlation between progressive
aspect and epistemic modality observed in Section 2 (Table 4), the German
cognate müssen does not exhibit a similarly rigid distinction: for example,
dynamic verbs may have a focal epistemic meaning and stative verbs a focal
agent-oriented meaning in the present tense, as the following examples show,
respectively:
(19) Er muß kommen; ich höre schon seine Stimme.
he must come I hear already his voice
'He must be coming; I hear his voice already.'
(20) Er muß heute nachmittag unbedingt zu Hause sein.
he must today afternoon [Link] at home be
'He definitely has to be at home this afternoon.'
Even in English, the distinction is not as clear-cut as Bybee, Perkins and
Pagliuca (1994) suggest. Consider an utterance like (21), where must may
have either an agent-oriented (21a) or an epistemic interpretation (21b)
(Perlmutter 1970:115; Calbert 1973:7); see also (26) below.
Agent-Oriented vs. Epistemic Modality 35

(21) Clyde must work hard.

a [A: "Clyde is obliged to work hard"]
b [E: "It must be the case that Clyde works hard"]
Second, with reference to property C, Coates (1983:245) observes that
the core meanings of agent-oriented modals in English are crucially associ­
ated with agentivity. This applies to German modals as well, as we saw in
Section 2, where action verbs, which typically require an agent, were shown
to trigger an agent-oriented interpretation of modals in the majority of exam­
ples. At the same time, however, agentivity is not decisive for defining agent-
oriented modality in German, as (17) illustrates. The same may be said for
property D, also demonstrated by (17), which contains neither a dynamic
predicate nor an agent-like participant. This raises the following question: If
agent-oriented modality can be defined with reference to one property only,
namely F, what accounts for its strong association with properties C, D, and L?
There appears to be a straightforward answer: since the presence of F
implies that an event may, should or must happen, this entails that the event
will occur later than now, and more often than not an event that happens will
involve agents and actions, less typically also inanimate forces. Thus, the
presence of F suggests that we are dealing with a kind of utterance that tends
to be associated with agents and dynamic processes typically leading to
changes of state. If, however, a modal utterance lacks F, i.e., if it expresses
epistemic modality, it is likely to be associated with states and time-stable
situations. It is not surprising therefore that epistemic modality is closely
associated with utterances involving existential subjects and Stative verbs, or
verbs in the progressive and perfect aspects.
That existential subjects, impersonal subjects and dummy subjects are
strongly associated with epistemic, rather than with agent-oriented modality,
has been argued repeatedly in the literature on English modality (Coates 1983;
Bybee, Perkins, and Pagliuca forthc). Bybee, Perkins, and Pagliuca (1994) in
fact observe that the presence of a dummy subject precludes an agent-oriented
reading in cases like (22):
(22) It should take me about four hours to get there.
While we have not been able to make corresponding observations about
German, there are contextual frames where the presence of a dummy subject
does favor an epistemic reading. Thus, while (23) is associated equally with
36 Bernd Heine

an agent-oriented and an epistemic interpretation, (24) may be said to have a

focal epistemic reading (24b), but only a non-focal agent-oriented reading
(24a).
(23) Der Brief muß in zwei Tagen da sein.
the letter must in two days there be
'The letter must be there in two days.'
(24) Es muß zwei Tage dauern, bis der Brief da ist.
it must two days last until the letter there is
'It must take two days until the letter is there.'
a. [A: "... I'll be in trouble if the letter arrives earlier"]
b. [E: "... since it is a long way to get there"]
Yet even in the presence of existential/dummy subjects and durative
verbs like dauern 'last', an agent-oriented sense is possible (24a). What might
account for the stronger epistemic force in the case of contextual frames like
(24) is the fact that the absence of a human or animate subject is also likely to
suggest absence of a manipulating force F and, hence, to favor an epistemic
interpretation. Note, however, that German has a paradigm instance of an
impersonal marker, which is man 'one', and this marker is almost exclusively
associated with agent-oriented modality. Thus, even in utterances with a main
verb kennen 'know' as their predicate, which strongly correlates with epis­
temic modality (as we saw in Section 2.2, Table 2), there is a focal agent-
oriented sense if the sentence subject is man, as in (25).

(25) Man muß sie kennen.

one must her know
'You have to know her.'
There appears to be an obvious reason for this: man only occurs as a
subject, refers exclusively to human referents, is associated with human
activities, and tends to imply some manipulating force. It is therefore not
surprising that it promotes agent-oriented, rather than epistemic, interpreta­
tions.
Agent-Oriented vs. Epistemic Modality 37

4. On models

4.1 Metaphor model vs. context model

A number of studies of grammaticalization have claimed that the develop­

ment of grammatical categories involves a metaphorical process (see espe­
cially Claudi and Heine 1986; Sweetser 1988; Heine, Claudi and Hiinnemeyer
1991; Stolz 1991; see also Bybee and Pagliuca 1985:73). For example, there
are good reasons to argue that the shift of the English be going to-construction
from a concrete/lexical meaning to the abstract/grammatical meaning of
future tense has a metaphorical base. First, this shift involves a transfer from
the "real-world", the world of referential entities and kinetic activities, to the
world of discourse, to entities that exist in acts of speech or, e.g. with
reference to the development of modal auxiliaries, a transfer from the socio-
physical world to the world of reason and belief (Sweetser 1984:24). Thus, we
are dealing with a transfer from one domain of human conceptualization to
another. Second, the metaphoric nature of this transfer can also be derived
from the fact that our be going to-example meets the criteria commonly
applied to define metaphor, e.g., that the statement, if taken literally, is false.
While in the sentence Sally is going to town the phrase is going to has its
literal meaning of motion in space, it has a "transferred sense" in sentences
like Sally is going to wake up in a minute, where the literal meaning is ruled
out.9
Essentially the same kind of metaphorical transfer can be observed in
other instances of grammaticalization: they involve a transfer from one cogni­
tive domain to another, with the effect that, from a certain stage onwards, an
expression undergoing this shift becomes false or meaningless if taken liter­
ally, that is, if interpreted as designating its original meaning. A number of
such source domains and their corresponding target domains are discussed in
Heine, Claudi and Hiinnemeyer (1991). It goes without saying that we are
dealing here with a specific kind of metaphorical transfer that has been
described as "emerging metaphor" (Heine, Claudi and Hiinnemeyer 1991:60-62).
There is, however, at least one problem with this metaphorical model.
This problem concerns the continuous nature of grammaticalization. Meta­
phor is commonly assumed to involve a discrete "jump" from one domain to
another - a process that is hard to reconcile with the gradient nature of
38 Bernd Heine

grammaticalization chains (Heine 1992). In the case of be going to, the shift
from physical motion to grammatical function proceeded in a chain-like
manner. This is suggested on the one hand by historical evidence (cf. Pérez
1990) and on the other hand, by the present use patterns of the construction.
That conceptual shift in the process of grammaticalization is chain-like
and continuous is due to the particular circumstances giving rise to new
grammatical meanings: such meanings are derived from existing meanings
via context extension (Aijmer 1985), and through the conventionalization of
invited inferences or conversational implicatures (Traugott and König 1991).
This reasoning has been advanced with reference to the transition from agent-
oriented to epistemic meanings. Thus, Traugott (1989:50-51) argues that from
'permission' one can implicate 'expectation': thus, if I say You may go, I may,
in the right circumstances, implicate that I want you to go, from which you
may infer that you have some obligation to go. With reference to the modality
of obligation, Traugott and König observe:
[...] must in the epistemic sense of 'I conclude that' derived from the
obligative sense of 'ought to' by strengthening of conversational inferences
and subjectification. If I say She must be married in the obligation sense, I
invite the inference that she will indeed get married. This inference is of
course epistemic, pertaining to a state of affairs that is anticipated to be true
at some later time (Traugott and König 1991:209).

This overall process, of which pragmatic strengthening appears to be a part, is

called 'context-induced reinterpretation' by Heine, Claudi and Hünnemeyer
(1991, Chapter 3), because it is specific contexts (or "circumstances") that
invite new inferences and ultimately lead to the emergence of (new) gram­
matical meanings.
To summarize, in order to account for the conceptual shift from concrete/
lexical to abstract/grammatical meaning we have two main models at our
disposal: one that rests on a metaphorical interpretation of the process and
may therefore be called the metaphor model, and another relying on context-
induced reinterpretation, which I call the context model. While in the past,
scholars have argued in favor of either one or the other, more recent investiga­
tions suggest that both models are required to understand the process con­
cerned.
There are, however, two contrasting positions with regard to the role
these models should play in a theory of grammaticalization. On the one hand,
there is the position of Traugott (1989), Traugott and König (1991), and of
Agent-Oriented vs. Epistemic Modality 39

Bybee, Perkins and Pagliuca (1994.), according to which different kinds of

grammaticalization have to be distinguished; and while some kinds are han­
dled best in terms of the metaphor model, others have to be accounted for in
terms of the context model. Among these authors again, there are differing
views as to where exactly the models apply. While all agree that the metaphor
model typically applies in the case of concrete/lexical structures and the
context model in the case of more strongly grammaticalized structures, disa­
greement exists as to where to draw the boundary between the two. For
Traugott (1989) and Traugott and König (1991), the transfer from concrete
source propositions to auxiliary functions of tense and aspect is metaphori­
cally structured11 while the transfer from agent-oriented (deontic) to epistemic
meanings is not:
But the shift to epistemics of conclusion, belief, knowledge, hear-say,
hypothetical conditionality, and so forth has little of the analogical mapping
from one conceptual domain onto another that is characteristic of metaphor
(Traugott 1989:50).

Sweetser (1982) on the other hand suggests that the conceptual shift from the
agent-oriented to the epistemic domain is a clear instance of a metaphorical
process. Bybee, Pagliuca and Perkins (forthc.) again argue that the English
modal auxiliaries must and should both have experienced an extension from
agent-oriented to epistemic modality, but whereas the extension of must is
suggestive of metaphorical transfer, that of should might be due to the
conventionalization of implicature and, hence, would have to be dealt with in
terms of the context model.
The different behavior of must and should had already been pointed out
by Coates (1983:14ff.). In her work on English modal auxiliaries, she ob­
served that in the transition from one kind of modality to another there are
three types of indeterminacy, which she refers to, respectively, as gradience,
ambiguity, and merger. Gradience concerns the nature of the continuum of
meaning, e.g., in the transition from the core of ABILITY to the periphery of
POSSIBILITY of English can. Ambiguity may be described in terms of the
following properties (cf. Coates 1983:15-16):

a. A given expression has two senses and it is not possible to decide

from the immediate context which of these is intended.
b. The two senses are in an either/or kind of indeterminacy since
either of them may be chosen.
40 Bernd Heine

c. These senses belong to different categories, viz. the categories of

agent-oriented (= root) and of epistemic modality, respectively.
d. The categories concerned are discrete.
The only example of ambiguity in Coates' work on English modal auxiliaries
involves must, which is ambiguous between an agent-oriented (26a) and an
epistemic sense (26b).
(26) He must understand that we mean business.
a. [A: "It is essential that he understand that we mean business"]
b. [E: "Surely he understands that we mean business"]
Merger shares with ambiguity properties (a) and (d). Instead of properties
(b) and (c), however, it has (b') and (c'), respectively:
b'. The two senses are in a both/and relationship, i.e., they are mutu­
ally compatible. In order to understand the relevant expression, it is
not necessary to decide which of the two senses is intended.
c'. In certain contexts, the two senses are "neutralized", i.e., they are
not mutually exclusive.
As examples of merger, Coates (1983:17) cites English should and ought to,
as in (27):
(27) a. A: Newcastle Brown is a jolly good beer.
B: Is it?
C: Well it ought to be at that price.
b. [A: Obligation of the maker to provide a good beer]
c. [E: Logical assumption - "the beer costs a lot, therefore it is
good"]
According to Heine, Claudi and Hünnemeyer (1991) on the other hand,
for all instances of grammaticalization analyzed so far, both the metaphor and
the context models are required, each relating to a different level: the context
model to the micro-level and the metaphor model to the macro-level of
conceptual shift. The former perspectivizes the continuous nature of concep­
tual shift, that is, the fact that this shift proceeds in a series of overlapping,
contextually defined extensions, while the macro-level perspectivizes the fact
that, given enough micro-level extensions, conceptual shift will cross bounda­
ries between cognitive domains, like that between the domain of concrete,
Agent-Oriented vs. Epistemic Modality 41

"real-world" phenomena and that of abstract grammatical functions. In the

case of the be goingto-examplealluded to above, the context model takes care
of the many micro-level extensions to be observed both in the historical and
the present uses of this construction, including the numerous situations of
ambiguity between physical motion and intention, or between intention and
prediction, or even among all three senses. The metaphor model, on the other
hand, is concerned with and accounts for the macro-level shift from a concrete
source proposition to an abstract grammatical function.12
The data considered in the previous sections suggest that the distinction
between agent-oriented and epistemic modality is of a discrete nature in the
case of German modals, being based on presence vs. absence of the modal
force F. It was noted, however, that three additional properties tend to be
associated with this distinction, namely presence vs. absence of a controlling
agent (C), of a dynamic event (D), and of the later-than-reference-time
constraint (L). The more of these properties are present, the more pronounced
the semantic contrast between the two interpretations is likely to be and the
more likely it is that we are dealing with instances of ambiguity; conversely,
the more of these properties that are absent, the more we approach what
Coates refers to as merger. Take, for example, utterance (13), repeated here as
(28):
(28) Er muß frieren.
he must freeze
'He is freezing/feeling cold.'
(They let him freeze.')
Muß in (28) is likely to involve a modal force F, hence, its focal sense is an
agent-oriented one, although apart from F, none of the properties normally
accompanying agent-oriented uses, i.e. C, D or L, is clearly present. There is,
however, an almost equally pronounced epistemic sense ["He must be freez­
ing"], and the two senses can be said to be suggestive of merger, in that they
satisfy criteria (b') and (c') mentioned above: given the right context, there is
not much of a difference whether the hearer interprets (28) in its agent-
oriented or in its epistemic sense. (28a) below would be an example of such a
context: the contextual information Er hat keinen Mantel an ("He isn't
wearing a coat") may be interpreted alternatively as the force that is responsi­
ble for the content expressed by (28) or else as a presupposition leading to the
logical assumption underlying the epistemic interpretation of (28).
42 Bernd Heine

(28) a. ["Er hat keinen Mantel an;..."]

'He isn't wearing a coat;...'

It is in utterances like (28), used in contexts like (28a), that an inferential

mechanism could have been at work (Bybee, Pagliuca and Perkins 1994)
whereby an agent-oriented sense of obligation gives rise to an epistemic sense
of inferred certainty.
What this would seem to suggest is that the transition from the agent-
oriented to the epistemic sense of (28) involves both discontinuity and conti­
nuity; discontinuity because even in a context like (28a), the two modalities
can clearly be distinguished, and continuity because, given the right context, it
does not matter which of the two senses is intended by the speaker and
selected by the hearer: the information value remains essentially the same. A
similar situation is found in the case of modals other than müssen, which
means that, for all modals considered, the same descriptive framework can be
employed. 13
To summarize, the fact that there is only one criterial property that
distinguishes the two kinds of modality in German modals would seem to
suggest, first, that an element of discreteness separates the two modalities.
Second, it suggests that the transition from agent-oriented to epistemic modal­
ity may be regarded as an instance of semantic bleaching in that it involves the
"bleaching out" of property F, i.e., of Jespersen's "element of will"
(1924:320-1). At the same time, however, there are also gains: with the
elimination of F, the modals acquire new senses that are characterized by what
Traugott (1989:49) calls the "strengthening of focus on knowledge, belief,
and the speaker's attitude toward the proposition".

Third, such observations suggest that there are reasons to regard the transition
from agent-oriented to epistemic concepts as being metaphorically structured,
as has been argued for independently by Sweetser (1982; 1988), Bybee and
Pagliuca (1985), and Heine, Claudi and Hünnemeyer (1991): it involves a
transfer from one domain of human experience to another, from the
sociophysical world to the epistemic world (Sweetser 1982:492-5), the former
being a dynamic world of willful human beings who act and are capable of
imposing their will on other agents, the latter essentially a static world, one
that may but need not be associated with human participants (Heine, Claudi
and Hiinnemeyer 1991:176-178).
Agent-Oriented vs. Epistemic Modality 43

4.2 Another example

In a paper on "Temporal distance: a basic linguistic metaphor", Fleischman

(1989) observes that distance in the past frequently serves to move a real event
from actuality to a hypothetical world, and that expressions for temporal
distance are often pressed into service to express, inter alia, social distance
and politeness, as in (29):
(29) / thought/was thinking about asking you to dinner.
In this way, Past tense forms referring to a present situation are "felt to be less
assertive, more deferential, whence more polite" than corresponding Present-
tense forms:
The risk of a possible negative reply is reduced since no invitation has
actually been issued. Use of the PAST tense in speech acts of this type
serves to cancel - or at least attenuate - their illocutionary force [...]
(Fleischman 1989:9).

The result is that temporal distance serves as a metaphorical vehicle for social/
interpersonal distance as well as for a number of other notions relating to
evidentiality, speaker subjectivity, etc.
A different explanation is volunteered by Bybee (This volume) with
regard to English modal verbs used in the Past tense. She discusses some of
the same evidence adduced by Fleischman but describes the phenomenon
without reference to metaphor; rather, her parameters are the particular se­
mantics characterizing modals on the one hand, and the use of past tense on
the other:
[...] the use of wanted to in present time in Modern English is possible
because it implies that certain conditions on carrying out the wanted predi­
cate may not be met. The so-called polite or remote uses of Past tense as in
(21) / wanted to ask you a question.
arise in the same way. That is, (21) implies that there might be conditions
that are unmet. Among these implied conditions is the question of whether
the addressee wants to be asked a question — thus the deferential use of the
Past. (Bybee, this volume).

According to Bybee, a series of factors work together to produce the

deferential or polite senses apparent in the two English examples presented
above:
44 Bernd Heine

a. First is the fact that modal verbs are stative verbs. When modals are
used in the Past what is asserted is that a state existed before the
moment of speech, but not that the state necessarily still exists in
the present.
b. Since the state may still exist in the present, a modal in the Past
tense may be used in contexts where it has present relevance: it
may receive the interpretation "that the modality is still in effect,
and [that] the predicate action will be carried out if the right
conditions are met" (Bybee, this volume).
c. The combination of the modal sense and the Past sense is said to
produce a hypothetical reading; compared to Present tense modals,
modals in the Past tense may then express hypothetical situations
in conditional apodosis, as can be seen in (30a), as opposed to
(30b):
(30) a. If I saw Judy, I would tell her the news.
b. If I see Judy, I will tell her the news.
d. In the context of a dialogic discourse (typically involving first and
second person referents), the hypothetical sense may then be ex­
ploited to express a deferential or polite sense, as in (29) above,
e.g., to tone down or weaken the illocutionary force of an otherwise
assertive speech act, as described in the above quotation by
Fleischman.
As is to be expected in such processes of grammaticalization, with the
development of new senses, old uses may gradually lose in significance. For
example, the more the hypothetical sense gains ground, the more the past
sense disappears. Stage (a) is no longer relevant for English Past tense modals
such as would and could', for should, Bybee (This volume) adds that "one
could argue that there are no past uses at all".
I do not wish to delve into the question as to whether the development
sketched in (a) through (d) is appropriate only to languages with a fuzzy/
overlapping zone between the morphological categories of past tense and
subjunctive, or to languages lacking a subjunctive category altogether. In
German, for example, Bybee's scenario does not seem to apply in cases where
there is a well marked formal distinction between Past tense and Subjunctive:
Agent-Oriented vs. Epistemic Modality 45

the role of a hypothetical or deferential sense is invariably assumed by the

Subjunctive rather than the Past tense, as can be seen in (31).
(31) Ich kann Ihnen helfen.
'I can help you.' PRESENT

Ich könnte Ihnen helfen. SUBJUNCTIVE

(Similar meaning, but more hypothetical, or polite, deferential)
Ich konnte Ihnen helfen.
'I was able to help you.' PAST
(Past significance only)

It would seem, however, that this question does not touch the main point made
by Bybee. What is obvious, is that Bybee's account offers a legitimate
alternative to that of Fleischman, one that rests on a series of inferences that
can be described in terms of context-induced reinterpretation (Heine, Claudi
and Hiinnemeyer 1991, Ch. 3).
Bybee (this volume) argues that Fleischman's metaphorical explanation
of a transfer from temporal distance to social distance in certain politeness
forms does not account for the fact that it is always modal verbs that are
involved in this transfer. Yet Bybee's account does not explain how it is that
modals which express a situation involving unmet conditions are also used for
marking a certain kind of social relationship (politeness). What I wish to argue
here again (cf. Heine, Claudi and Hiinnemeyer 1991), is that an explanation
for a conceptual shift in the process of grammaticalization must have two
components: one relates to the macro-effects of the process which are sugges­
tive of a shift from one domain of human conceptualization to another, in this
case from the domain of time to that of social relations; the other relates to the
micro-effects which are suggestive of a gradient sequence of contextual and
inferential extensions leading to the emergence of chain-like use patterns of
the linguistic items concerned.
While Fleischman's analysis highlights the macro-level of the process,
Bybee's approach perspectivizes the micro-level. The way in which the two
models are interrelated is sketched graphically in Figure 1 (cf. Heine, Claudi
and Hiinnemeyer 1991, Ch. 4). Thus, rather than contradicting or excluding
one another, the two models complement each other and are both required for
a better understanding of grammaticalization.
46 Bernd Heine

Domain
TIME SOCIAL RELATIONS

Context

Figure 1. A metaphor-and-context model for some of the senses of English would and
should.

5. Conclusions

In previous research it has been established that epistemic uses of modals tend
to develop out of agent-oriented uses; that is, the latter are older than the
former (see especially Shepherd 1982; Bybee and Pagliuca 1985; Traugott
1989; Bybee, Pagliuca and Perkins 1994.). From this it follows that contexts
associated with agent-oriented modality reflect more conservative or less
grammaticalized use patterns than contexts associated with epistemic modal­
ity. While there may be exceptions, this conclusion seems to hold true for
most of the data considered here; it is also confirmed by diachronic analysis.14
With reference to German this would mean in particular that the most con­
servative behavior is encountered if
a. modals such as möchten 'would like to', dürfen 'be allowed to,
may', or wollen 'want', rather than modals like sollen 'should' or
können 'can' are involved,
b. the main verb is an action verb like kommen 'come',
c. the modal is used in the perfect or the past tense,
d. the modal occurs in interrogative rather than in declarative utter­
ances,
e. the modal occurs in negative rather than in affirmative utterances, or
f. if the subject is a first or second person, rather than a third person
referent.
Agent-Oriented vs. Epistemic Modality 47

Some related observations have also been made for English modals. Traugott
(1989:52), for example, notes that older meanings tend to be maintained
longer in negative environments, and draws attention to the relative paucity of
epistemic (as opposed to agent-oriented) uses of must not and mustn't in
British English and to the maintenance of the volitional sense of will in We
won't go.
One main claim made in this paper is that it is not possible to describe the
transition between agent-oriented and epistemic modality in German modals
exclusively in terms of linguistic categorization. What has to be taken into
consideration in addition are the contextual frames with which particular uses
of a modal are associated. Utterances having human agents as their subject
and dynamic main verbs may still have a focal epistemic sense if associated
with a contextual frame that discourages an agent-oriented interpretation.
More important, however, is the presence vs. absence of the modal force F in
determining which variety of modality a given modal utterance receives.
Finally, as we saw in Section 4, the transition from one kind of modality
to another has elements of discontinuity as well as continuity, and can thus be
accounted for by appeal to both the metaphor model and the context model of
grammaticalization. In this respect, the problem looked at in this paper does
not differ significantly from other instances of grammaticalization studied so far.

NOTES

1 I wish to express my gratitude to Werner Abraham, Joan Bybee, Eithne Carlin, Wallace
Chafe, Ulrike Claudi, Zygmunt Frajzyngier, Tom Givón, Christa Kilian-Hatz, Christa
König, Andreas "Donald" Lessau, Frank Lichtenberk, Franz Potyka, Heinz Roberg,
Dan Slobin, Thomas Stolz, Eve Sweetser, as well as a number of other participants of the
Albuquerque symposium on Mood and Modality for valuable discussions on this paper,
most of all to Suzanne Fleischman. I am also indebted to the Deutsche Forschungs­
gemeinschaft (German Research Society) for its financial support.
2 In the following, y will use the terms "agent-oriented" and "deontic" as defined by
Bybee, Pagliuca and Perkins (1991, 1994), with slight modifications to be specified
below. Note that the distinctions referred to above are not only terminological; for
Coates (1983), for example, "root modality" embraces "deontic" and "dynamic catego­
ries", and "epistemic modality" also includes "alethic" modality.
3 See Steele et al. (1981:260-264) and Heine (1993) for additional exemplification.
4. Note, however, that, compared to full-fledged verbs, they have a reduced paradigm of
personal inflections, lacking e.g. the present tense 3sg. suffix -t; see below.
 48 Bernd Heine

5 Square brackets "[]" are used in this paper for contextual information relevant for the
semantic interpretation of an utterance.
6 While it is hoped that the observations made in the course of the survey will reveal some
salient semantic characteristics of German modals, the survey is biased in several ways.
For example, the three informants, all students of the University of Cologne having some
acquaintance with linguistics, had to take the three-fold classification proposed here for
granted in their responses.
7. The situation would be different if subjunctive uses of modals were included.
8 I am referring to the non-standard progressive of the form Er ist am Essen 'He is eating'
which, although not acceptable in High German, is nevertheless found in some form or
other in many modern German dialects.
9 Cf. the notion of preference as defined by Givón (1990:529).
10. For further evidence on the metaphorical nature of conceptual shift in grammaticaliza-
tion, see Heine, Claudi and Hünnemeyer 1991.
11 The examples provided by Traugott (1989:207) involve such developments as that from
GO to future, from COME to perfect, or from BE AT/IN to progressive. The process
concerned is described by Traugott as one involving "spatio-temporal metaphors".
12 In addition to these two models, a third model has to be distinguished, namely the
bleaching or containment model (Tom Givón, p.c.; see Heine, Claudi and Hiinnemeyer
1991:108ff.). We will not further deal with this model here, essentially since it is
confined to semantic substance while our primary concern is with the cognitive and
pragmatic strategies underlying the process of grammaticalization (but see below).
13 As noted above, this does not necessarily apply to werden, whose status as a modal is controversial.
14 The present paper is based exclusively on synchronic findings.

REFERENCES

Abraham, Werner 1989. "Futur-Typologie in den germanischen Sprachen." Abraham and

Janssen 1989. Pp. 345-369.
Abraham, Werner 1992. The Aspectual Source of the Epistemic-Root Distinction of
Modal Verbs. Paper presented at the Symposium on Mood and Modality,
Albuquerque, University of New Mexico, May 8-10, 1992.
Abraham, Werner and Theo Janssen 1989. Tempus - Aspekt - Modus: Die lexikalischen
und grammatischen Formen in den germanischen Sprachen. Tübingen: Max
Niemeyer Verlag.
Ahlqvist, Anders ed. 1982. Paper from the Fifth International Conference on Historical
Linguistics. Amsterdam: John Benjamins.
Aijmer, Karin 1985. The semantic development of will. Fisiak 1985. Pp. 11-21.
Bybee, Joan L. and William Pagliuca 1985. Cross Linguistic Comparison and the Devel­
opment of Grammatical Meaning. Fisiak, J. 1985. Pp. 59-83.
Bybee, Joan L., William Pagliuca, and Revere D. Perkins 1991. Back to the Future.
Traugott, Elizabeth C. and Bernd Heine 1991.2. Pp. 17-58.
Agent-Oriented vs. Epistemic Modality 49

Bybee, Joan L., Revere D. Perkins, and William Pagliuca forthc. The Grammaticization of
Tense, Aspect, and Modality in the Languages of the World. Chicago, London:
University of Chicago Press.
Bybee, Joan L. This volume. The Semantic Development of Past Tense Modals in
English. Paper prepared for the Symposium on Mood and Modality, University of
New Mexico, Albuquerque, 8 - 1 0 May, 1992.
Calbert, Joseph P. 1971. "Modality and Case Grammar." Working Papers in Linguistics
(Ohio State University), August 1971:85-132.
Claudi, Ulrike and Bernd Heine 1986. "On the Metaphorical Base of Grammar." Studies
in Language 10,2:297-335
Coates, Jennifer 1983. The Semantics of the Modal Auxiliaries. London/Canberra: Croom
Helm.
Fisiak, Jacek, ed. 1985. Historical Semantics, Historical Word Formation. Berlin:
Mouton de Gruyter.
Fleischman, Suzanne 1989. "Temporal Distance: A Basic Linguistic Metaphor." Studies
in Language 13,1:1-50.
Givón, T. 1990. Syntax: A Functional-Typological Introduction. Volume II. Amsterdam/
Philadelphia: John Benjamins.
Haiman, John 1985. Iconicity in Syntax. (Typological Studies in Language, 6.) Amster­
dam/Philadelphia: John Benjamins.
Heine, Bernd 1992. "Grammaticalization Chains." Studies in Language 16,2:335-68.
Heine, Bernd 1993. Auxiliaries. New York: Oxford University Press.
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer 1991. Grammaticalization: A
Conceptual Framework. Chicago: University of Chicago Press.
Helbig, Gerhard and Joachim Buscha 1988. Deutsche Grammatik. Ein Handbuch für den
Ausländerunterricht. Eleventh edition. Leipzig: VEB Enzyklopädie.
Jacobs, R. and P. Rosenbaum, eds 1970. Readings in English Transformational Gram­
mar. Waltham, Mass.: Ginn and Co.
Jenkins, L. 1972. Modality in English Syntax. Indiana University Linguistics Club:
Mimeograph.
Jespersen, Otto 1924. The philosophy of grammar. London: Allen and Unwin.
Lakoff, George & Mark Johnson 1980. Metaphors We Live by. Chicago/London: Univer­
sity of Chicago Press.
Langacker, Ronald W. 1985. "Observations and Speculations on subjectivity." In Haiman
1985. Pp. 109-150.
Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Pérez, Aveline 1990. "Time in Motion: Grammaticalisation of the Be Going to Construc­
tion in English." La Trobe University Working Papers in Linguistics 3:49-64.
Perlmutter, D. M. 1970. "The Two Verbs 'Begin'." In Jacobs, R. and P. Rosenbaum 1970.
Pp. 107-119.
Ramat, Anna Giacalone 1991. "Grammaticalization Processes in the Area of Temporal
and Modal Relations." To appear Studies in Second Language Acquisition 14 (1992).
Shepherd, Susan C. 1982. "From Deontic to Epistemic: An Analysis of Modals in the
History of English, Creoles, and Language Acquisition." In: Ahlqvist, Anders 1982.
Pp. 316-323.
 50 Bernd Heine

Steele, Susan M., Adrian Akmajian, Richard Demers, Eloise Jelinek, Chisato Kitagawa,
Richard Oehrle, and Thomas Wasow 1981. An Encyclopedia of AUX: A study in
Cross-Linguistic Equivalence. (Linguistic Inquiry Monographs, 5.) Cambridge,
Mass./London: MIT Press.
Stephany, Ursula 1989. "Modality in First Language Acquisition: The State of the Art."
To appear in Proceedings of the Symposium on Modality in Language Acquisition,
Berlin, Free University, 1-3 May, 1989.
Stolz, Thomas 1991. Von der Grammatikalisierbarkeit des Körpers. I: Vorbereitung.
Typescript, University of Bochum.
Sweetser, Eve Eliot 1982. Root and Epistemic Modals: Causality in Two Worlds."
Berkeley Linguistics Society 8: 484-507.
Sweetser, Eve Eliot 1984. Semantic Structure and Semantic Change: A Cognitive Linguis­
tic Study of Modality, Perception, Speech Acts, and Logical Relations. Ph. D. disserta­
tion, University of California at Berkeley.
Sweetser, Eve Eliot 1988. Grammaticalization and Semantic Bleaching. Berkeley Lin­
guistics Society 14:389-405.
Traugott, Elizabeth C. 1989. "On the Rise of Epistemic Meanings in English: An Example
of Subjectification in Semantic Change." Language 65,1:31-55.
Traugott, Elizabeth C. & Bernd Heine, eds 1991.1, 1991.2. Approaches to Grammatical­
ization. 2 volumes. Amsterdam/Philadelphia: John Benjamins.
Traugott, Elizabeth C. and Ekkehard König 1991. "The Semantics-Pragmatics of Gram­
maticalization Revisited." Traugott, E. C. and B. Heine 1991.1. Pp. 189-218.
Vater, Heinz 1970. On the Generation of Modal Verbs. Paper read at the Ethno-Linguistic
Seminar, Bloomington, Indiana.

ABBREVIATIONS

A/E = focal agent-oriented/epistemic sense

Al/El = non-focal agent-oriented/epistemic sense
A2/E2 = marginal agent-oriented/epistemic sense

APPENDIX

Agent-oriented vs. epistemic modality of eight German modals

The data presented below have been elicited from three native speakers of High German.
While a number of discrepancies were noted among these informants, the overall ratios
turned out to be essentially the same.
Agent-Oriented vs. Epistemic Modality 51

A. Linguistic data

1. Present
Er muß kommen A E2 Er muß Geld haben E Er muß sie kennen E
Er soll kommen A E Er soll G. haben E Er soll sie kennen E
Er kann kommen A E2 Er kann G. haben A E Er kann sie kennen E
Er will kommen A Er will G. haben A E Er will sie kennen E
Er mag kommen A E2 Er mag G. haben A2 E Er mag sie kennen E
Er darf kommen A Er darf G. haben A Er darf s. kennen
Er möchte kommen A Er möchte G haben A Er möchte s. kennen A
Er wird kommen A E2 Er wird G. haben A2 E Er wird sie kennen A2 E

2. Past
Er mußte kommen,. A Er mußte Geld haben. Er mußte sie kernten.
sollte A E sollte E sollte
konnte A konnte A konnte
wollte A wollte A wollte E2
mochte A mochte A2 mochte
durfte A durfte A durfte A

3. Perfect
hat kommen Er hat Geld haben Er hat ihn kennen
müssen. A müssen. A2 müssen. A2
sollen. A sollen. A2 sollen. A2
können. A können. können. A2
wollen. A wollen. A wollen. A2

4. Negation
Er muß nicht kommen. A Er muß kein Geld haben. Er muß sie nicht kennen. A2 E
soll A soll A soll Al E
kann A kann Al E kann E
will A will A will A E
mag A mag A2 mag A El
darf A darf A darf A
möchte A möchte A möchte A
wird A wird Al E wird E
52 Bernd Heine

5. Question
Muß er kommen? A Muß er Geld haben ? A2 Muß er sie kennen? A E
Soll A Soll A1 Soll
Kann A Kann A El Kann E
Will A Will A Will El
Mag A Mag Al Mag
Darf A Darf A Darf A
Möchte A Möchte A Möchte A
Wird A Wird Al Wird E2

6. Main verb in perfect

Er muß gekommen sein. E Er muß Geld gehabt haben. E Er muß sie gekannt haben. E
soll E soll E soll E
kann E kann El kann E
will E will E will E
mag E mag E mag E
darf darf darf A2
möchte möchte E2 möchte A2
wird E wird E wird E

7. Second person subject

Du mußt kommen. A Du mußt Geld haben. E Du mußt sie kennen. E
sollst A sollst E sollst E
kannst A kannst A E2 kannst E2
willst A willst A E2 willst E2
magst magst E2 magst E
darfst A darfst A darfst A2
möchtest möchtest A möchtest A2
wirst A1 wirst Al E wirst E

8. First person subject

Ich muß kommen. A Ich muß Geld haben. A Ich muß sie kennen. E
soll A soll E2 soll E
kann A kann A1 kann El
will A will A will A2
mag A mag E2 mag E
darf A darf A2 darf Al
möchte A möchte A möchte A
werde A werde A werde E
Agent-Oriented vs. Epistemic Modality 53

0 = minimal value)
Index of modality
(1.00- maximal,
B Quantitative data (Based on A above)

modality
Value of

Maximum number of points

Number of instances
i

J
J

Agent-oriented
Agent-oriented

Epistemic

Epistemic
Kind of proposition

1 Present (3rd person subject) 41 43 24 72 0.57 0.60

2 Past 33 10 18 54 0.61 0.19
3 Perfect 21 0 12 36 0.58 0
4 Negation 56 23 24 72 0.78 0.32
5 Question 52 11 24 72 0.72 0.15
6 Main verb in perf. 2 54 24 72 0.03 0.75
7 2nd p. subject 33 26 24 72 0.46 0.36
8 1st p. subject 39 16 24 72 0.54 0.22

Kind of expression

Er kommt. 137 28 58 174 0.79 0.16

Er hat Geld. 94 61 58 174 0.54 0.35
Er kennt sie. 46 94 58 174 0.26 0.54

Kind of modal

Dürfen 47 0 7 63 0.75 0
Möchten 41 1 6 54 0.76 0.18
Wollen 48 23 8 72 0.67 0.32
Können 38 30 8 72 0.53 0.42
Müssen 31 31 8 72 0.43 0.43
Mögen 23 25 7 63 0.37 0.40
Werden 25 32 6 54 0.46 0.59
Sollen 30 40 8 72 0.42 0.56
 The Expression of Root and Epistemic
Possibility in English1

Jennifer Coates
Roehampton Institute

1. The root/epistemic distinction

The distinction between root (or agent-oriented or deontic) modality and

epistemic modality has proved enormously useful to those attempting to
describe the modal systems obtaining in the world's languages. The analysis
of English is no exception (see Coates 1983; Haegeman 1983; Palmer 1990).
Moreover, there is general agreement on the definition of these terms. Epis­
temic modality is concerned with the speaker's assumptions or assessment of
possibilities, and in most cases it indicates the speaker's confidence or lack of
confidence in the truth of the proposition expressed. Root modality encom­
passes meanings such as permission and obligation, and also possibility and
necessity. This means that the root/epistemic distinction cuts across the neces­
sity/obligation and possibility/permission distinctions (see diagram below).
As in many languages, in English the same linguistic forms express both root
and epistemic meanings. So, for example, may can express both root and
epistemic possibility; must can express both root and epistemic necessity.

Root Epistemic

CAN permission possibility possibility MAY

MAY

MUST obligation necessity necessity MUST

HAVE TO ← → HAVE TO

Figure 1. Meaning and the root/epistemic distinction

56 Jennifer Coates

As far as the expression of necessity is concerned, this polysemy is unprob-

lematic: the root/epistemic distinction remains distinct, as the following ex­
amples illustrate. Must, in (la), expresses the root meaning of obligation, in
(lb) it expresses the (weaker) root meaning of necessity, and in (lc) it
expresses epistemic necessity.

(1) a. You must finish this before dinner ROOT

b. All students must obtain the consent of the Dean ROOT
c. I must have a temperature EPISTEMIC

It is my impression, however, that there is some confusion about the root/

epistemic distinction when it is applied to possibility. In this paper, I want to
re-examine this area of modality, and will argue that: (i) the semantic contrast
between root and epistemic possibility is considerably weaker than in other
root/epistemic pairs; (ii) where the same linguistic form expresses both root
and epistemic possibility, instances of merger (see Coates 1983:17) are
common (may); (iii) where a linguistic form expresses predominantly root
possibility, epistemic readings are likely to develop (can).

2. Root and epistemic possibility

In order to discuss the root/epistemic contrast in relation to the expression of

possibility in English, I shall make use of the set of properties developed by
Heine (this volume) in his analysis of the German modals. Heine argues that
the following properties are criterial:
a. There is some force F that has an interest in an event either occurring or
not occurring (Heine comments that F "is characterised by some 'ele­
ment of will', to use the wording of Jespersen (1924:320-1)". (Heine this
volume)
b. That event is to be performed by some agent A.
c. The event is dynamic (D).
d. The event has not yet taken place at reference time and, if it does take
place, that will be at a time later than reference time (L).
e. The event is non-factual, but there is a certain degree of probability that it
will occur (P).
The Expression of Root and Epistemic Possibility in English 57

In an utterance such as she must go to bed now, F may refer to the speaker
(T insist that she goes to bed now') or to an absent parent ('her mother insists
that she goes to bed at this time') or to any other source of power the speaker
may have in mind. The pronoun she is the agent (A), the event — go — is
dynamic (D), the event has not yet taken place at the moment the utterance is
produced (L), and there is a high probability (P) that the event referred to will
take place.
As Heine (this volume) says: "While prototypical instances of agent-
oriented modality are characterised by the presence of the properties [i.e. F,
A, D, L and P], prototypical instances of epistemic modality lack all proper­
ties except P". This means that the fewer of these properties there are in any
given instance, the weaker will be the semantic contrast between the two
interpretations of these sentences. If we look at the examples given in (1)
above, we can justify the claim that the semantic contrast between root and
epistemic meaning is strong here by applying Heine's criteria. In (la) and (lb)
all of the properties are present, while only P is present in (lc). In other words,
examples of root must in English would normally be classified as prototypical
examples of agent-oriented modality, while examples of epistemic must
would be classified as prototypically epistemic.
Let's turn to the contrast between root and epistemic possibility. Examples
(2) and (3) below are typical instances of the expression of root possibility:2
(2) well I think there is a place where I can get a cheap kettle
(S. 1.4.62)
(3) I am afraid this is the bank's final word. I tell you so that you may
make arrangements elsewhere. (W.7.9.37)
Can (see example 2) is the normal exponent of root possibility in English,
while may, as illustrated in example (3), is the exponent of root possibility in
more formal contexts (in this case, a letter from a Bank Manager). Properties
A, D, and L are present in both (2) and (3), but F (force) is absent.
In everyday discourse, can expressing root possibility is most commonly
found in examples like (4) and (5) below, general statements of possibilities
with impersonal subjects.
(4) certain things can be sex-linked to the Y chromosome (S.5b.2.54)
(5) first thing in the morning they come, you can hear the whistle
(S.1.14A.43)
 58 Jennifer Coates

These two examples display none of Heine's properties apart from P. Even
more perplexing are archetypal examples of can such as (6) below, which
Palmer (1990:152-4) describes as 'existential':
(6) Lions can be dangerous
Palmer justifies his use of the term 'existential' by claiming that paraphrases
involving 'some' {some lions are dangerous) or 'sometimes' {lions are some­
times dangerous) are more appropriate than paraphrases using 'possible for'.
What is intriguing about examples like (6) is that, besides lacking F, A, D, and
L (the properties associated with non-epistemic modality), they also lack P. In
other words, some examples of can lack all the properties which Heine claims
are normally associated with modal meaning.
Although we have seen that examples of root possibility vary in the
number of properties associated with them (examples (2) and (3) are associ­
ated with properties A, D, L and P, examples (4) and (5) are associated only
with P, while example (6) is associated with none of the properties), they all
share one characteristic: absence of F. In his analysis of German modals,
Heine argues that only F is obligatorily connected with root modality: "once
the use of a modal is characterised by a lack of F then it lacks that 'element of
will' that Jespersen (1924:320-1) referred to, and we are dealing with an
epistemic rather than an agent-oriented sense" (Heine, this volume). On the
basis of this argument, (2), (3), (4), (5) and (6) all fail as instances of root
modality.
Let's look now at the expression of epistemic possibility in English.
Epistemic possibility has many exponents, notably maybe, perhaps, I think,
possibly, probably and the modal auxiliaries may, might and could. Examples
are given in (7), (8), (9) and (10) below.

(7) that may be yellow fever, I'm not sure (S.4.2.65)

(8) I may be a few minutes late, but don't know (S.7.3E.6)
(9) / think it's unlikely actually but he might do it today (S.8.1A.18)
(10) The only snag is that it has been raining ... and I could get held up
for anything up to a week. (W.7.2.29)
These examples are not associated with F. However, examples (8), (9) and
(10) are associated with L (since they refer to events that are to occur later
than reference time), and example (9) is also associated with A and D. Thus,
only (7) is prototypically epistemic in Heine's sense.
The Expression of Root and Epistemic Possibility in English 59

Example Properties Type of modality

F A D L P

la + + + + + root obligation
1b + + + + root necessity
lc + epistemic necessity
2/3 - + + + + root possibility
4/5 +
6 - - - - - 'existential'
7
8
9
-
+ +
+
+ + epistemic possibility

Figure 2.
+
Matrix showing presence or absence of Heine's properties

A summary of the properties associated with examples given so far is

presented in the figure above. This shows that, using Heine's criteria alone,
we are unable to distinguish clearly between instances of root and epistemic
possibility in English. This figure includes examples (la), (lb) and (lc) to
show how clearly Heine's properties distinguish between root and epistemic
necessity. But in the case of examples of root and epistemic possibility in
English, this table would force us to conclude that there is often no difference
(cf. examples (2) and (9) which share four properties; examples (4) and (7)
which both have one property, P). While the use of these criteria has helped to
demonstrate that the distinction between root and epistemic possibility is
weak, it is not the case that examples are indistinguishable. So what distin­
guishes root from epistemic possibility?
The crucial distinction between forms expressing root possibility in
English and forms expressing epistemic possibility in English is that the latter
involve Subjectivity (which I shall refer to as S). Forms involving S can be
defined as 'devices whereby the speaker, in making an utterance, simultane­
ously comments upon that utterance and expresses his attitude to what he is
saying' (Lyons 1977:739). As examples (7), (8), (9) and (10) above illustrate,
S is an integral component of the expression of epistemic possibility. The
speakers in (7), (8), and (9) and the letter-writer in (10) are not only making
statements but are simultaneously expressing their lack of confidence in the
propositions expressed in these utterances. In every case here, speaker's
uncertainty is encoded in an accompanying phrase which reinforces the
 60 Jennifer Coates

modal: I'm not sure in (7), don't know in (8), I think it's unlikely in (9), and the
only snag is in (10). (We can compare earlier examples of root possibility: I can
get a cheap kettle (2) and you can hear the whistle (5). Both are statements of
fact; subjectivity is not involved. The difference between I can get a kettle and I
may get a kettle is that in the latter the speaker's uncertainty is encoded too.)
We therefore need to add S to the matrix to show that this acts as the
criterial property where the expression of possibility is concerned, (see Figure 3)
If we ignore P, which is common to both root and epistemic modality
(except in unusual cases like (6)), we can see that the reason for the weakness
of the root/epistemic distinction in the expression of possibility is the absence,
in utterances involving root possibility, of properties normally associated with
root meaning. Exponents of root necessity (e.g. lb) differ from exponents of
epistemic necessity (e.g. lc) on five measures, whereas exponents of root
possibility (e.g. 2/3) may differ from exponents of epistemic possibility (e.g.
9) by as little as one property (presence or absence of S).
One of the reasons that the weakness of this distinction has been unprob-
lematic for speakers of English is that root and epistemic possibility are, by
and large, expressed by different linguistic forms. May is the only modal form
which regularly expresses both root and epistemic possibility, and when
expressing root possibility it is restricted to the most formal contexts (as in (3)
above). The extent of overlap between the two forms is small, as the following
statistics show (these record my analysis of a representative sample of 200
cases of can and 200 cases of may, all examples of spoken British English
taken from the Survey of English Usage):

Example Properties Type of modality

F A D L P S

la + + + + + - root obligation
lb + + + + + - root necessity
2/3 — + + + +
root possibility
4/5 + :}
6 - - - - - - 'existential'

lc - - - - + + epistemic necessity
7 - - - - +
8
9
-
-
-
+
-
+
+
+
+
+ + epistemic possibility

+
Figure 3. Matrix to show the distinction between root and epistemic meanings in English

+
 The Expression of Root and Epistemic Possibility in English 61

Table 1. The use of can and may in contemporary spoken English

can: Root Possibility 129 may. Epistemic Possibility 147

Ability 41 Permission 32
Permission 10 Root Possibility 7
Undecidable 20 Valediction 1
TOTAL 200 Undecidable 13
TOTAL 200

As these figures demonstrate, each linguistic form is overwhelmingly associ­

ated with one particular meaning {can with root possibility, may with epis­
temic possibility). The category 'undecidable' refers to those examples which
I could not assign unproblematically to one meaning rather than another.
Often, this was because there was not enough contextual evidence to permit a
definite reading; in the case of may, however, some examples exhibit
'merger', that is, the phenomenon whereby utterances containing may com­
bine elements of both root and epistemic meaning (see 3.1 below).

3. The linguistic consequences of a weak root/epistemic contrast

In the previous section, I have demonstrated the weakness of the root/epis­

temic distinction in the expression of possibility. As I have argued, the
blurring of the root/epistemic boundary in this semantic area is not often
problematic in English, given the use of different linguistic forms. However,
there are certain interesting linguistic developments in this area which are a
direct result, I would argue, of the weakness of the distinction. One involves
the growing instances of merger with may. The other is the development of
epistemic readings for can in contemporary spoken American English.

3.1 May and merger

Merger, as I have explained elsewhere (see Coates 1983; Leech and Coates
1980), refers to instances where two meanings co-exist in a both/and relation­
ship. In other words, two readings are available for a given utterance, but
instead of having to choose one meaning and discard the other (as with
ambiguous examples), the hearer is able to process both meanings. Merger
occurs quite frequently in more formal texts (example 11) and is becoming
endemic in academic writing (as example (12) illustrates):
62 Jennifer Coates

(11) or the pollen may be taken from the stamens of one rose and
transferred to the stigma of another (W. 10.3.27)
(12) ... the process of simplification ... through which even forms and
distinctions present in all the contributory dialects may be lost
(Trudgill 1986:126)
In both these examples, the only property clearly present is P. Properties
associated with root meaning (F, A, and D) are absent, while S, normally
criterial for epistemic meaning, is not typically associated with this kind of
formal style with its passives and inanimate subjects. So instances such as
these lack clear markers of either root or epistemic meaning. The two mean­
ings merge, and the reader is not required to choose one or the other:3

(11') Root: 'it's possible for the pollen to be taken ...'

Epistemic: 'it's possible that the pollen will be taken ...'
(12') Root: 'it's possible for forms and distinctions ... to be lost'
Epistemic: 'it's possible that forms and distinctions ... will be lost'
It is significant that such examples of merger come typically from the
(formal) written domain. In spoken language, exponents of root and epistemic
possibility are normally distinguished prosodically: for example, may and can
do not receive stress when expressing root possibility, but may, when it
expresses epistemic possibility, is normally stressed. Epistemic may is also
typically associated with fall-rise intonation (see example (13) below):
(13) I may be wrong (S1.2.38)
While writers can exploit the both/and relationship of root and epistemic
meanings when talking about possibilities, speakers are constrained by pro­
sodic factors to choose one or the other. If a speaker uttered example (11) with
a fall-rise nucleus on may, for example, the only interpretation available to
hearers would be epistemic.

3.2 The development of epistemic can

Given the historical pattern of epistemic meanings developing from root

meanings (Bybee & Pagliucca 1985; Traugott 1989), it would not be surpris­
ing if can were to develop an epistemic reading. Moreover, the homogeneous
picture of can given so far in this paper is not the whole truth: can does have
some well-established specialised epistemic uses.
The Expression of Root and Epistemic Possibility in English 63

First, it provides the missing negative form in the epistemic must para­
digm (see Palmer, this volume). The invariant form can't (not cannot or can
not) expresses 'it's necessarily the case that ... not ...' (nec ~ p) or 'it's not
possibly the case that' (~ poss p) in examples such as (14). (Note the stress on
can't and the fall-rise intonation contour.)
(14) [speaker describes friends arriving early]
I almost phoned them up and said come a bit later — and then I
thought oh they 've probably left by now — so I didn 't and — twelve
thirty, now that... can't be them, and it was (S.2.7.6)
Second, can is used in interrogative constructions to express epistemic
possibility. Example (15) is a cliché of pop music and Hollywood-style films,
but it makes the point clear:
(15) Can it be true?
['Is it possible that this is true? that she loves me?']
Such examples have a clear relationship with the use of can't discussed above,
in that (15) could be glossed it can't be true! (it must be false). Example (16)
comes from a radio discussion of Government policy on pensions:
(16) Can that be sensible ?
['Is it possibly the case that that is sensible?']
The speaker was clearly trying to make the point That can't be sensible!
In British English, these are still the only contexts in which a form of can
is used with epistemic meaning (though could is making headway as an
alternative to might in the expression of tentativeness — see example 10). But
in American English, can is starting to appear in other contexts. The following
example occurred during the Symposium on Mood and Modality (held at the
University of New Mexico in May, 1992) as a participant finished her presen­
tation:
(17) we hope this coding system can be useful [to other linguists work­
ing in the field]
This utterance meant something like 'we hope there's a chance that this
system will be useful'. For British speakers, this utterance is not possible: a
British speaker would have to say 'we hope this coding system will be useful',
thereby losing the subjective force. As the person who uttered (17) com-
64 Jennifer Coates

mented to me afterwards, can in this utterance is 'a sort of hedge'. American

speakers seem to have no problems with utterances of this kind.
So if can is developing an epistemic meaning in the United States but not
in Britain, what is different in the linguistic environment in the United States?
I would like to suggest three factors:
a. can is less commonly used to express permission in American English;
may is the normal exponent of permission (Coates & Leech 1980);
b. the 'bleaching' of root can is further advanced in American English, with
the majority of examples not associated with F, A, D or L;
c. may is the chief exponent of epistemic possibility in British English, but
is less common in American English, where may has connotations of
formality.
More generally, there are many reasons why we would expect can to
develop epistemic possibility readings. First, all the other modal auxiliaries in
English express both root and epistemic meaning. Second, it seems to be the
case that epistemic meanings derive from earlier, non-epistemic meanings
(Traugott 1989:52). Third, the evidence from child acquisition research is
suggestive: children develop deontic meaning much earlier than epistemic
meaning (Stephany 1986); Guo (this volume) claims that the Mandarin form
neng {can) is starting to be used to express epistemic possibility in children's
speech. Fourth, the occurrence of merger with examples of may (illustrated in
Section 3.1) illustrates the fuzziness of the root/epistemic boundary in the
expression of possibility. Fifth, as this paper has attempted to demonstrate,
root and epistemic possibility are only weakly distinguished.
Given these circumstances, and a historic pattern in the development of
can from ability meanings to permission meanings to root possibility mean­
ings (Bybee 1988), I would predict that initially examples of epistemic can
will co-occur with syntactic features such as inanimate subject and stative
verb, and in contexts where accompanying words support an epistemic read­
ing. Example (17) is of this type, with an inanimate subject this coding system,
and the phrase I hope introducing subjectivity to the utterance. However, it is
difficult to imagine can becoming a serious contender in the expression of
epistemic possibility unless it can develop a stressed alternative to the usual
[kən]; until that happens, utterances like I can come will be processed as root.
The Expression of Root and Epistemic Possibility in English 65

4. Conclusions

In this paper, I have argued that the distinction between root and epistemic
meanings is much weaker in the case of possibility than in other areas of
modal meaning. I have demonstrated that the weakness of this distinction
arises from the nature of root possibility, typical examples of which are not
associated with Heine's properties F, A, D and L (normally criterial for root
meaning). As a direct consequence of the weakness of the root/epistemic
distinction, instances of merger are common (in the case of may), and epis­
temic readings are beginning to occur in declaratives with can. Speakers will
exploit the potentialities of the English modal system to say the things they
need to say. Whether this means that can, like the other English modal
auxiliaries, will develop the full range of epistemic meanings remains to be
seen.

NOTES

1 I would like to record my gratitude to Bas Aarts, Joan Bybee, Suzanne Fleischmann,
Talmy Givón, Bernd Heine and Charles Meyer for their comments on earlier drafts of
this paper.
2 Examples from this point onwards will be taken from the Survey of English Usage,
University College, London (prosodic information omitted).
3 Can is also beginning to be involved in merger, in contexts where the speaker/writer
wishes to hedge what they are saying. A nice example occurs on Inland Revenue (UK)
Tax Forms: False statements can result in prosecution.

REFERENCES

Bybee, Joan. 1988. "Semantic Substance Versus Contrast in the Development of Gram­
matical Meaning". Proceedings of the Fourteenth Berkeley Linguistic Society 247-279.
Bybee, Joan & William Pagliuca. 1985. "Cross-Linguistic Comparisons and the Develop­
ment of Grammatical Meaning". Historical Semantics and Historical Word Formation
ed. by J. Fisiak, 59-84. The Hague: Mouton.
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries, London: Croom Helm.
Coates, Jennifer & Geoffrey Leech. 1980. "The Meanings of the Modals in Modern
British and American English". York Papers in Linguistics 8.23-34.
Guo, Jiansheng. This volume. "The Interactional Stucturing of Meaning: Children's Use
and Development of the Mandarin Modal neng 'can'".
66 Jennifer Coates

Haegeman, Liliane. 1983. The Semantics of Will in Present-day British English. Brussels:
Paleis der Academiën.
Heine, Bernd. This volume. "Agent-Oriented vs. Epistemic Modality — Some Observa­
tions on German Modals".
Jespersen, Otto. 1924. The Philosophy of Grammar. London: Allen & Unwin.
Leech, Geoffrey & Jennifer Coates. 1980. "Semantic Indeterminacy and the Modals".
Studies in English Linguistics ed. by Sidney Greenbaum, Geoffrey Leech & Jan
Svartvik, 79-90. London: Longman.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Palmer, Frank. 1990. Modality and the English Modals. (2nd edition). London: Longman.
Stephany, Ursula. 1986. "Modality". Language Acquisition ed. by Paul Fletcher &
Michael Garman. (2nd edition) 375-400. Cambridge: Cambridge University Press.
Traugott, Elizabeth Close. 1989. "On the Rise of Epistemic Meanings in English: An
Example of Subjectification in Semantic Change". Language 65.31-55.
Trudgill, Peter. 1986. Dialects in Context. Oxford: Blackwell.
Contextual Conditions for the Interpretation
of 'poder' and 'deber' in Spanish*

Carmen Silva-Corvalán
University of Southern California

1. Introduction

This study deals with some aspects of the grammar of poder 'can, may' and
deber 'must' in Spanish. Poder and deber are considered to be modal verbs or
modal auxiliaries both on syntactic and semantic grounds (see, among others,
Marcos Marín 1975:211; Narbona 1989; Rivero 1977).1 Many grammars of
Spanish (e.g. Gili Gaya 1976; Hernández 1986; Real Academia Española
1973) include querer 'want', saber 'know', osar 'dare', soler 'be used to', etc.
as 'modal verbs' as well. Syntactically, however, poder, deber, osar and soler
are different in that they share the requirement to occur exclusively in con­
struction with an Infinitive with an equivalent subject, a constraint which does
not apply to querer and saber (see Narbona 1989, and Rivero 1977 for further
discussion of the syntax and semantics of these and other periphrastic verbal
constructions). Semantically, poder and deber stand apart as the only modals
that can make a statement, in at least one interpretation, about the possibility
of p, as the paraphrases of examples (1 a-b), given in (2 a-b), show.

(1) a. Jon puede/debe tocar el clarinete.

b. Jon quiere/sabe/osa/suele tocar el clarinete.
a. 'Jon may/must play the clarinet'
b. 'Jon wants to/knows how to/dares to/is in the habit of playing
the clarinet'
(2) a. It is possible/very likely that Jon plays the clarinet.
b. It is the case that Jon wants/knows how to/dares to play/often
plays the clarinet.
68 Carmen Silva-Corvalán

There are, then, syntactic and semantic arguments for treating poder and
deber as a class in Spanish. In contrast to modal verbs in English, whose
syntactic and semantic characteristics have been studied in depth (see Coates
1983, Perkins 1982 & 1983, Palmer 1977 & 1979, Halliday 1970, and numer­
ous references in these studies and in the present volume), Spanish modals
have received little attention, perhaps due to the fact that their syntactic
characteristics have not appeared to be as distinct as those of the English
modals when compared with other verbs which may occur in auxiliary posi­
tion in verbal periphrases.
Note, however, that poder and deber pose a number of special syntactic
and semantic questions of general and specific interest: What syntactic con­
straints apply to these modal verbs? What is the most adequate way to
describe their semantics? Can a monosemantic approach account for the
various uses oí poder and deber? What is the relation between the messages
conveyed by poder and by deber (as in examples 3-4)? What role do the
linguistic and extra-linguistic contexts play in the interpretation of the mean­
ing of these forms? Indeed, as can be seen in the suggested readings of the
following examples, poder and deber can convey different messages, which
include such notions as ability, permission, possibility, and necessity.

(3) Juan puede venir.

'John can/may come.'
Possible paraphrases:
a. John is able to come.
b. John is allowed to come.
c. It is possible for John to come/that John will come.
(4) Juan debe venir.
'John must come.'
Possible paraphrases'.
a. John has the obligation to come.
b. It is very likely/I can confidently infer that John will come.
In addition, poder and deber cannot be interpreted to convey possibility
in their non-finite forms, as the unacceptabihty of the paraphrases in (5 b) and
(6 b) indicate.
 Interpretation of 'poder' and 'deber' 69

(5) a. Juan va a poder cruzar el río.

'John's going to be able/be allowed to cross the river.'
Unacceptable reading'.
b. *Va a ser posible que Juan cruce el río.
'It may be the case that John will cross the river'
(6) a. Juan va a deber cruzar el río.
'John's going to have to cross the river.'
Unacceptable reading:
b. *Es muy posible que Juan vaya a cruzar el río
'It's very likely that John's going to cross the river'
By contrast, when combined with conditional verb morphology, poder is
interpreted to convey possibility (ex. 7), while deber can only convey (weak)
obligation, as in example (8).
(7) Juan podría hacerlo.
'John could do it'
Possible reading'.
It may be possible for John to do it.
(8) Juan debería hacerlo.
'John should do it'
Unacceptable reading:
*'It's very likely that John would do it'
In this essay I examine the factors that determine the different interpreta­
tions of poder and deber in different contexts, and argue for the adequacy of a
monosemantic approach in the analysis of the semantics of these verbs. I
propose to show that the 'modal verbs' each have an invariant meaning (IM),
but are interpreted to convey different contextual meanings (CM) as a conse­
quence of their interaction with other elements in the context (e.g. animacy of
NPs, aspect, tense).
I approach the problem mainly through an examination of examples of
poder and deber in actual language use.2 My analysis suggests that a monose­
mantic approach is more appropriate for Spanish on at least two grounds: (1)
The different meanings proposed (ability, possibility, permission, politeness)
are shown to be a function of the interaction between the modal and other
70 Carmen Silva-Corvalán

elements (linguistic and extra-linguistic) in the discourse. (2) A monoseman-

tic perspective, which admits the possibility of synonymy in context, can
account for the choice of one "synonymous expression" over another on the
basis of their different basic meanings. The invariant meanings proposed for
poder and for deber justify the choice of one modal over the other in a
particular utterance where both may be interpreted to convey, for instance,
epistemic possibility (more strictly 'epistemic necessity' in the case of deber,
but I will not make a difference between possibility and necessity here). While
the analysis of the data allows me to argue against a polysemantic approach to
the meaning of modals, it also leads me to propose that their contexts of
occurrence may be polysemantic, i.e., at times more than one interpretation of
the meaning of a given modalized utterance is possible. Thus, the inclusion of
a specific context (defined by a combination of features, e.g. animacy and
agentivity of entities) in one or another set of contexts associated with the
inference of different messages (e.g. possibility, permission) is in some cases
fuzzy.

2. Theoretical frameworks

Linguistic studies of modality in natural languages differ at least in three

respects: (1) the definition and types of modality identified; (2) the establish­
ment of what constitutes 'true' modality; and (3) the theoretical approach used
in accounting for the semantics of modality.
Thus, Palmer (1979), for instance, uses modality to refer to the meanings
of the English modal verbs, classified on the basis of their formal features. By
contrast, Halliday (1970) considers modalities to be only those meanings
related to the assessment of probability. A modality is expressed by a form
which represents "the speaker's assessment of the probability of what he is
saying, or the extent to which he regards it as self-evident" (p. 328). There­
fore, though all modal verbs in English may express modalities, in Halliday's
view they are not modality when they express permission, volition, ability,
obligation, etc.3
Bybee (1985:169) explicitly establishes a more sensible and appropriate
difference between mood, a verbal category which has also been considered to
represent the speaker's assessment of the degree of factuality of the proposi­
tion, and modality in the following terms: modality is "a conceptual domain
[including root and epistemic meanings] which may take various types of
Interpretation of 'poder' and 'deber' 71

linguistic expression," while mood refers to a subdivision of this conceptual

domain and is realized as "a marker on the verb that signals how the speaker
chooses to put the proposition into the discourse context." (165) That is,
mood, a grammatical category signalled by verbal affixes, is one among
several means of expressing modalities.
Different theoretical models have been adopted in studies of modality in
English (Coates 1983; Haegeman 1983; Kratzer 1977; Perkins 1983; Sweetser
1982). Haegeman, Kratzer and Perkins propose a basic or core-meaning
approach to account for the meaning of modals,4 arguing that this core
meaning is present in all their uses and is compatible in different contexts with
the various notions which have been associated with the modals (e.g. neces­
sity, permission). Out of context, however, i.e. in their lexical entry form, only
the basic meaning is present.
Coates rejects both a purely monosemantic and a purely polysemantic
approach: the former, she insists, fails to deal with the problem of indetermi­
nacy in real language, while the latter has to assume discrete categories of
meaning in a semantic field characterized precisely by indeterminacy. In­
stead, she seeks to reconcile the two approaches observing, on the one hand,
that the root/epistemic distinction is discrete,5 but on the other, that examples
assigned to the two categories are indeterminate inasmuch as they cover a
range of possible meanings. Quite naturally, then, Coates directs her attention
to 'fuzzy set' theory (Zadeh 1965, 1972), a model which allows her to account
for clearly classifiable discrete cases as belonging to the core of the set, as
well as for various degrees of indeterminacy or gradience as belonging to the
skirt or the periphery.
In contrast, Sweetser views the 'possibility' use of modals as a meta­
phorical extension of root meaning. Adopting the 'force dynamics' semantics
of Talmy (1982, cited in Sweetser 1982), she argues that the reasoning
processes underlying epistemic meanings are "subject to compulsions, obliga­
tions, and other modalities, just as our real-world actions are subject to
modalities of the same sort" (484), i.e. to those modalities (ability, necessity,
permission) considered to convey root or deontic meanings. Accordingly, an
understanding of the epistemic world of possibilities and logical conclusions
in terms of the sociophysical world explains quite naturally the connection
between, for instance, epistemic may as the absence of a barrier to the
speaker's conclusions about the truth of the proposition, and permission
-granting (i.e. root) may as the absence of a potential barrier for the truth of the
proposition in the sociophysical world.
72 Carmen Silva-Corvalán

Sweetser further agrees with the observation made by other linguists (e.g.
Coates, Kratzer, Perkins) that pragmatic factors will determine which world
the modal is taken as operating in. Thus, the identification of a real-world
cause will determine a root reading, while the identification of a body of
premises will determine an epistemic one.6 In either case, the contextual
interpretation of such modals as can and may (as well as of verbs like let and
allow) appears to be that of "taking away a potential barrier" (485). Sweetser
acknowledges that there may be ambiguity between real-world force and
epistemic force, which in turn implies contextual ambiguity, but she leaves
open the question of the basic meaning or basic contribution that different
modals make to the interpretation of an utterance.
Like Coates, I feel that "the imprecision of our knowledge of the world
might be inherent" ([Link]), and that fuzzy set theory may be of use in the
analysis of modality (cf. Sweetser on contextual ambiguity). Contrary to
Coates (and perhaps also to Sweetser), however, I argue that modals have
basic, core meanings along the lines of the definitions proposed by Perkins
(1983), while fuzziness applies rather to the contexts, linguistic and extra
-linguistic, in which modals are used. Fuzziness or graded membership of
contexts allows for various interpretations of modalized propositions, and the
precise point at which one interpretation is no longer possible is usually
difficult to establish. Contextually inferred messages (e.g. ability, possibility),
in principle infinite, have been mistakenly considered to be the, or part of the,
meaning of the modal.7
The question of the basic meanings conveyed by linguistic expressions,
as opposed to the meanings, messages or pragmatic implicatures which these
forms appear to have in specific contexts of use, is a recurrent theme in
semantic-pragmatic studies. With particular respect to closed grammatical
systems (e.g. pronouns, verbal affixes), there seems to be general agreement
that a distinction between basic, invariant, or systemic meaning, and second­
ary, implicated, or non-systemic meanings or functions must be made (Bello
1977; Bull 1971; García 1975; King 1992; Silva-Corvalán 1991). By contrast,
there is little agreement as to what exactly constitutes the semantic substance
of the linguistic units under analysis.
With respect to poder and deber, I propose that these modals may be
characterized as having invariant meanings which account for the use of the
modal instead of a non-modalized verb, or for the choice of one modal over
the other in particular discourse contexts. Poder and deber interact with other
Interpretation of 'poder' and 'deber' 73

linguistic and extra-linguistic elements that contribute to producing the multi­

ple sentential and/or discourse meanings that are compatible with the seman­
tics of the modals.
I am suggesting, then, the existence of three meaning components or types
of meaning in language: (1) de-contextualized, systemic invariant meaning;
(2) contextualized meaning; and (3) prototypical discourse meaning.8 These
meaning components are not specific to modals, but characterize all linguistic
elements; contextualized and prototypical meanings are related to decontextu-
alized invariant meanings in systematic, principled ways which need to be
empirically ascertained.
Invariant meaning (IM) refers to the meaning which underlies, or is
present in, all uses of a modal; contextualized meaning (CM) is the message
which the modal conveys, or the analyst infers that the modal conveys in a
specific context. Contextualized meanings derive from the interaction of the
modal verb with morphosyntactic, semantic, prosodie, and pragmatic factors
which constitute its context of use. Thus, every linguistic element contributes
a specific meaning to the total, but the CM goes beyond the sum of these
meanings because it incorporates pragmatic factors that may not be context
-independent as the IMs of grammatical forms (as opposed to lexical forms)
appear to be. Prototypical discourse meaning (PM) refers to the most frequent
message (contextualized meaning) that the modal conveys in a corpus of
language data.
A PM tends to correspond to the meaning that most language users (and
sometimes even linguists) assign to a form. This confusion arises from the fact
that the inferred message is so frequent that speakers assume it is part of the
meaning of the form (cf. Faltz 1989, cited in Bybee, Perkins & Pagliuca 1994:
Ch. 6).
IMs and CMs correspond in part to what Bosch (1985) calls 'context
independent lexical meaning' and 'context dependent contextual notions',
respectively. Contextual notions are in principle infinite and unique for each
new context, an observation which leads Bosch to propose that "we should
give up the notion of the meaning or interpretation of a sentence or utterance
as an identifiable unit or thing altogether" (257). By contrast, it seems to me
that there is reason to propose an intermediate level of contextual meanings
which are associated with or derived from classes or types of contexts, rather
along the lines of what García (1975:276-7) discusses in terms of "the chief
uses of the meaning" of a form or "great classes of messages", all of which are
74 Carmen Silva-Corvalán

characterized by the same common reason for using one form rather than
another. Thus, though there is in principle no end to the messages that may be
postulated, I would like to propose that CMs may cluster into "great classes of
messages" (the most frequent ones becoming the PMs) that underlie groups of
utterances where the modal is inferred to convey the same or similar meanings.
The term 'modality', as the expression of speaker attitude towards the
contents of a proposition, is a semantic notion covering a range of meanings:
certainty, probability, possibility, belief, obligation, necessity, permission,
volition, intention, doubt, prediction, and denial. The speaker's attitude may
be conveyed through verb morphology (mood), lexically, syntactically, pro-
sodically, or by a combination of these linguistic devices. Here, Ï focus on
lexical and affixal manifestations of epistemic and root modalities. For exam­
ple, the proposition Juan viene mañana 'John is coming tomorrow' can be
modalized to express speaker uncertainty or prediction about the probability
of actualization of the event encoded in the proposition (henceforth p) by
using the modal verbs (henceforth m) poder or deber. Thus, Juan puede/debe
venir mañana, may be paraphrased, for instance, as 'Speaker believes that it is
possible for Juan to come/that it is very likely that Juan will come tomorrow'.
Marked with Future morphology, on the other hand, these modals may not
convey possibility, but rather dynamic or deontic modalities: Juan podrá/
deberá venir mañana 'Juan will be able to/will have to come tomorrow'.
I distinguish between epistemic (EP) and root possibility (RP) contextual
meanings (cf. Coates 1983). Epistemic and root possibility are concerned with
the speaker's assessment of or assumptions about the likelihood that the
content of a proposition is or may become true. EP and RP differ in that the
former involves the speaker in logical inference and has p and m in its scope,
while the latter has only p in its scope and is agent/event oriented, i.e., the
subject of the modalized infinitive is normally a willful agent and/or the
infinitive is a dynamic verb.9 This difference has consequences for negation,
such that negation affects p in the case of epistemic sense, but m and p in the
case of root sense. When an affirmative modal (example 9) interpreted to
convey RP is negated, the negative form is interpreted as a negative fact
(example 10).10 To preserve the possibility interpretation, therefore, root
modality must be negated on p (example 11), while epistemic modality may
be negated on m, as example (12) shows.
Interpretation of 'poder' and 'deber' 75

(9) Ahí pueden fumar. Root possibility

'They may smoke there'
Paraphrase'.
It's possible for them to smoke there.
(10) Ahí no pueden fumar. Permission negated.
'They can't smoke there.'
Paraphrase'.
It is the case that they are not allowed to smoke there.
(11) Ahí pueden no fumar. Epistemic possibility
'They may not smoke there.'
Paraphrase:
It is possible that they do not smoke there.
(12) Juan no debe estar en casa. Epistemic possibility11
'John must not be home.'
Paraphrase'.
It is very likely/I confidently infer that John is not home.
I further use the term root modality to encompass both the deontic and
dynamic categories of modal logic, namely obligation, necessity, permission
(see examples 13-14), and ability (example 15), noting that though ability is
not usually considered to be truly a modality, it is relatable to deontic modal­
ity, and at times may be indistinguishable from RP, as Bybee, Perkins &
Pagliuca (1994) note to be the case in a large number of languages.
(13) Debes comer para sobrevivir. Obligation
'You must eat to survive.'
(14) [Father to young son]
Puedes hablar cuando yo me calle. Permission
'You can speak when I stop talking.'
(15) Juan puede andar en bicicleta; aprendió de pequeño. Ability
'John can ride a bicycle; he learnt when he was a child.'
The invariant meanings (IM) for poder and deber that I propose here are
approximately those that Perkins (1982) postulates for the English modals can
and must.12 The definition of modals is formulated with variables ranging over
different systems of laws and principles and different sets of circumstances or
76 Carmen Silva-Corvalán

contexts of use. It incorporates four elements: 1) the invariant or basic

meaning of the modal, which relates the variables C and X (defined below); 2)
K, representing one of various systems of laws or principles (e.g. social,
magic, natural, rational laws, which embody our knowledge of the world,
including those principles which regulate linguistic interaction) according to
which the modalized proposition can be interpreted; (3) X, the third element,
represents the event, state of affairs, etc. referred to in the proposition; and (4)
C, which stands for a set of circumstances, presupposed or explicitly identi­
fied, under which K is relevant.
The circumstances include the structure of p, which may consist of an
n-place predicate (y,z), where y = subject, z = predicate. This interpretation of
C differs from that of Perkins, for whom C is a variable which may represent
such notions as 'personal interactions', 'deontic source', 'subject-oriented'.
The problem is that it is not clear how Perkins's definition accounts for the
role of linguistic characteristics of p, including mood for instance, in the
interpretation of the meaning of modalized p.
The various elements that bear upon the interpretation of a modal are
summarized in the following formula: K (C..[IM]..X), which I illustrate
below. The ensuing sections study usesofpoder and deber and focus on C and
X in order to examine the interaction between these variables and the specific
semantic contribution proposed for the modals.

3. Poder

I assume that the formula proposed by Perkins (1982) for can captures the
meaning of poder. K (C does not preclude X). This formula incorporates both
the IM and the elements that contribute to the CM. Thus, poder presents the
IM 'does not preclude' and it relates actualization of X to a set of circum­
stances C, such that with reference to a set of principles K, the circumstances
C do not preclude X. The values of K and C are recognized (and specified by
the analyst) in context; these variables (which include morphosyntactic, se­
mantic and prosodic features) plus the modal chosen determine whether X is
interpreted under a dynamic, deontic or epistemic modality (Heine (this
volume) proposes similar variables in his stimulating account of German
modals).
Interestingly, the specific semantic contribution proposed for poder,
'does not preclude', appears to agree with the meaning postulated for can by
Interpretation of 'poder' and 'deber' 11

Bybee, Perkins & Pagliuca (1994: Ch. 6), i.e., 'enabling conditions exist', as
well as with the meaning suggested by Sweetser (1982) in terms of the
presence or absence of 'barriers'. The question might arise why I agree with
Perkins in assigning a negative meaning to poder. This is because it seems to
me that a negative meaning captures more appropriately than a positive
meaning (such as 'enables' or 'makes possible') the sense of 'difficulty
overcome' which appears to be incorporated in most of the CMs of poder.
This is taken up again later in this essay.
Example 16 illustrates how the definition of poder would lead to an
interpretation of ability assuming certain values for K and C. Different values
for K and C would obviously lead to different interpretations.

(16) Mi hija puede contar hasta diez.

'My daughter can count up to ten'
IM: K (C does not preclude X)
(i) K: natural/biological laws (e.g. learning capacity)
(ii) C: (a) speaker's previous experience of p (empirical circumstan­
ces, i.e. speaker's evidence that p is true)
(b) y = agent ('my daughter'); z = event ('count up to 10');
time orientation: generic (i.e. valid for an unspecified past,
present and future time)
(iii) X: my daughter counts up to 10

CM: given humans' learning capacity and the circumstances in C, ad­

dressee interprets that speaker intends to convey the message that y (his/her
daughter) has the ability for z ('count up to ten').
Given an affirmative sentence, the elements in C that appear to bear upon
the CM include (cf. Heine, this volume):
(i) the presence or absence of a deontic source,
(ii) evidentiality or previous experience of p (cf. Givón 1982, Wright
1990),
(iii) degree of agentivity of the subject,
(iv) aspect of the situation (stative, dynamic, etc.),
(v) morphological aspect and mood,
such that various combinations of these elements account for the meanings
traditionally associated with poder, 'ability, permission, possibility', i.e. these
so-called meanings are indeed contextual inferences.
78 Carmen Silva-Corvalán

Pottier (1976) discusses the same three meanings for pouvoir 'can' in
terms of 'pouvoir physique/possibilité interne', 'pouvoir par autorisation/
possibilité externe', and 'probabilité/possibilité neutre' (pp.39-41). In addi­
tion, Palmer (1977:2) observes that epistemic modality should be seen as a
more basic concept "since what is (epistemically) possible includes what can
(dynamically) be done". This led me to think, for a while, that the IM of poder
should explicitly include 'possibility'. It became obvious to me later, how­
ever, that 'possibility' is a CM which may be derived from 'does not pre­
clude', since not precluding X in fact implies (though not equivalently)
making X possible.

3.1 Poder in Present tense form.

The data examined indicate that examples of poder in Present tense affirma­
tive form group into five sets of CMs (or "great classes of messages"),
discussed below, with the following readings: permission, ability, mitigation,
root possibility, and epistemic possibility. It is in principle possible, however,
that other sets of CMs might be identified in a much larger corpus of data.
It must be emphasized that many cases are not easy to fit neatly into one
of these five CMs. The indeterminacy of many cases may be viewed as
stemming from possible indeterminacies of K and C in natural language use,
e.g. the strength of the deontic source, weak indication of evidentiality.
Analogous observations made by Coates in regard to the meaning of English
modals led her to develop a prototype model of analysis. As in her English
data, my data contain a number of examples with contexts which appear to
represent the core set of features associated with a certain CM, but many
examples fall somewhere in the skirt or towards the periphery of the set.
Let us look now at examples of poder in contexts whose features have led
me to propose five different modalities (sets of CMs).

3.1.1 Permission.
Whenever a strong deontic source, i.e. some person or institution that clearly
creates obligation or permission (Lyons 1977:843), is identified either explic­
itly or implicitly, the CM inferred is permission, as in example (14). The most
clearly identifiable context for permission, then, includes a strong deontic
source, an agentive subject, and future time orientation. The transcribed
Interpretation of 'poder' and 'deber' 79

conversations examined do not offer these clearly identifiable contexts. This

is most likely due to the nature of the data, which do not contain pragmatic
situations that would elicit the granting of permission. Example (17) is repre­
sentative: it seems to assume a deontic source, but its non-specific agent and
the generic rather than future time orientation make our interpretation move in
the direction of root possibility.
(17) En España a los dieciocho años ya puedes votar, (M18,p34) 13
'In Spain you can vote when you're eighteen.'
(i) K: social laws (governments have authority over citizens)
(ii) C: (a) deontic source implied (legal code of laws)
(b) y = agent; z = event; time orientation: generic
(iii) X: y,z
CM: y is allowed to z
(Possible CM: it is possible for y to z)
Both deontic source and set of laws under which a specific example is to
be interpreted may lead to indeterminacy. On the one hand, written codes of
laws (e.g. traffic regulations, state laws) constitute strong deontic sources and
contribute to a more clearly identifiable CM of permission. On the other hand,
ethical principles may be somewhat weaker sources of obligation, and a
parent to a child, or a teacher to a student, may or may not constitute a strong
deontic source. Weaker deontic sources (e.g. people at the same authority
level, abstract entities, such natural phenomena as river currents) determine
contextual meanings which move away from permission towards interpreta­
tions of ability or possibility, as in example (18).

(18) Creo que ésta es una democracia y que uno puede hacer lo que
quiera. (M25,p.6)
T think that this is a democracy and one can do what one wants.'
In example (18) the deontic source is an abstract entity, 'democracy' (as
opposed to Franco's dictatorship), and X is not a specific event. In addition,
time orientation is generic. Given these values for C and X, example (18) may
lend itself to an interpretation of root possibility as well as permission.
That these inferred meanings are contextually determined is sup­
ported by the fact that a change in the environment of the modal has conse­
quences on its interpretation, as shown in (19).
80 Carmen Silva-Corvalán

(19) [Father responding to son's request for permission to go out]

Confío en ti, puedes hacer lo que quieras.
T trust you, you can/may do what you want.'
A specific animate deontic source, a specific agent, and the pragmatics of
the interaction (request for permission) make the permission CM of poder
much less peripheral.

3.1.2 Ability.
The IM 'does not preclude' is interpreted as 'ability' when the following
contextual factors obtain: (a) K: natural or rational principles that permit us to
infer either innate or learned capacities to do z. (b) C: (i) the speaker has
evidence that X has been previously actualized, or that at least z has been, and
the circumstances are such that z may be valid also for a 'new' y; (ii) y is
animate, agentive, and specific; z is an event; core time orientation is generic
or present; if y is 'new', it is future.
Examples (20) and (21) illustrate contextual conditions for the CM 'ability':
(20) Ella os puede decir todo, todo referente a la droga. (M1 l,p39)
'She can tell you everything, everything about drugs.'
(21) El árabe y el ruso, pues, no los ha llegado a hablar, pero vamos,
para conversación puede defenderse. (M26)
'Arabic and Russian, well, she hasn't managed to speak them, but,
in conversation she can get by.'
Note that saber 'to know' may also convey 'ability for z', as in Yo sé
nadar 'I know how to swim'. Saber, however, simply asserts knowledge or
ability for; it does not implicate precluding circumstances as poder does.
Therefore, saber is not compatible with circumstances that imply that X is
somewhat surprising or that it involves effort. Thus, when y performs an event
that involves effort, or the overcoming of a given barrier, or a certain degree
of difficulty, p may be modalized with poder (ex. 22) but not with saber
(example. 23):
(22) Juan puede nadar mariposa por horas.
'John can swim the butterfly (stroke) for hours.'
(23) *Juan sabe nadar mariposa por horas.
*'John knows how to swim the butterfly (stroke) for hours.'
Interpretation of 'poder' and 'deber' 81

Further evidence for the implication of 'overcoming difficulty' is provided by

a comparison of example (21) with its non-modalized version in (24).
(24) El árabe y el ruso, pues, no los ha llegado a hablar, pero vamos,
para conversación se defiende.
'Arabic and Russian, well, she hasn't managed to speak them,
but,in conversation she gets by.'
Without poder, example (24) says that the subject referent (the speaker's
daughter) does well in conversation, a statement which is too assertive pre­
cisely because it does not convey the idea that this person encounters a certain
amount of difficulty in speaking Arabic and Russian. Example (25) is also
illustrative.
(25) No, no claro, vamos, de hecho esas voces que se graban, tú con tu
oído no las escuchas...solamente las puedes escuchar cuando
rebobinas la cinta y escuchas lo que se grabó. (M5,585)
'No, no, well, in fact some of those voices which you record you
can't hear with your naked ear...you can only hear them when you
rewind the tape and listen to what was recorded.'
From example (25) one infers that a certain amount of difficulty has to be
overcome for the modalized event to take place. In fact, the speaker explicitly
states that only after rewinding the tape can certain voices be heard. The
implication of the existence of precluding cirsumstances derived from the IM
of poder weakens the strength of the assertion.

3.1.3 Mitigation.

By weakening the assertiveness of utterances, the use oí poder may be said to

convey a CM of mitigation, which may at times be interpreted as politeness.
This is clearly seen in utterances containing a verbum dicendi, especially with
a first person subject. The fixed conversational expression puedo/podemos
decir 'I/we might say', shown in example (26), constitutes a typical example
of mitigated assertiveness.
(26) Pero en la vida social, podemos decir, no nota diferencias con, con
las gentes que están en cualquier barrio madrileño. (M26,212)
'But in social life, we might say, you don't notice differences
among people who live in any barrio in Madrid.'
82 Carmen Silva-Corvalán

3.1.4 Root possibility.

The clearest context for the interpretation of root possibility includes (a) a
non-agentive, non-volitional, non-specific subject (y); (b) a stative verb (i.e.,
z is not an event); (c) no evidence of previous actualization of p; (d) non-past
time orientation. Thus, the only CM that may be inferred for examples (27) &
(28) is RP.
(27) Ya puede tener todos los millones que quiera [la persona sigue
siendo buena, si es buena por naturaleza]. (M9,p.7)
'(S)he may have all the millions (s)he wants [the person continues
to be good, if (s)he's good by nature].'
(28) Una música diferente, a, digamos, a la música que tú
(non-specific) puedes escuchar de un piano, de un violin, etc.
(M5,663)
'Music which is different, shall we say, from the music you
(non-specific) might hear from a piano, a violin, etc.'
Many examples, however, with animate subjects and more or less dy­
namic verbs may be interpreted to convey both ability and RP (ex. 29), or to
be closer to an ability reading (ex. 30a) or to an RP interpretation (ex. 31).
(29) Argentina puede salir muy bien del paso, en cinco o seis años.
(M9,A464)
'Argentina can/may very well solve its problems, in five or six
years'
(30) No se dan cuenta que salen estos militares y creo xx, entonces
como son unos bestias, pues pueden xx unas matanzas (a) y pueden
hacer cosas mhm tremendas. (M10,pl3)
'They don't realize that these military men go out into the streets
and I think xx, then since they are like animals, well they can xx
killings (a) and they can do mhmm terrible things.'
(31) El colegio está cerquísima. Puede ir andando. (M1 l,p9)
'The school is very near. You may/can walk there'

3.1.5 Epistemic possibility.

EP involves the speaker's logical inference and lack of confidence about the
possibility of actualization of p. There are a number of clear examples of EP
interpretation in the data examined. In these clear cases, C includes one or
Interpretation of 'poder' and 'deber' 83

more of the following elements: (a) inanimate, non-specific subjects, or

obligatory subjectless sentences; (b) a stative verb (e.g. ser 'to be', estar 'to
be', querer 'to want', saber 'to know'); (c) negative consequences of z for y.
Further, when a negated Infinitive is modalized by affirmative poder, an
epistemic sense must be inferred (as in example 11). Examples (32-33)
illustrate.

(32) ...que si hoy se cierra el plazo, dentro de tres meses ya puede ser el
examen. (M5,135)
'...if the deadline is today, then the exam may be within three
months.'
CM: Based on previous experience, given that today is the deadline to
register for the exam, I infer that it is possible that the exam may be
given in three months.
(33) Puede ser ésa una diferencia respecto a otros regímenes tota­
litarios. (M5,250)
'That may be a difference with respect to other totalitarian re­
gimes.'
In the case of epistemic possibility, K includes rational laws (people can
make predictions, deduce, infer, etc.). As for C, lack of evidentiality and
stative aspect are indeed crucial. If evidentiality exists, an example may then
be interpreted as conveying RP, as in (34) and (35):
(34) No es un clima muy estándar. Efectivamente, te puede llover y al
día siguiente hacer un sol espléndido. (M6,p2)
'It's not an even climate. In fact, it can rain one day and the
following day the sun may be shining brightly'
CM'. I have evidence that the climate is not even. I may infer with
confidence that it is possible for it to rain one day and the following
day be sunny.
(35) Luisa puede ser la Decana de Filosofía, porque ha renunciado a su
otro cargo.
'Louise can be the Dean of Philosophy because she has resigned
her other job.'
CM: Now that she has resigned her other job, I infer that it is possible for
Louise to be the Dean of Philosophy.
84 Carmen Silva- Corvalán

Example (35) appears to be indeterminate between a RP and an ability

CM. This is to be expected when the context does not offer core contextual
conditions because it seems rather clear that CMs are all related through the
assumption of ability for z existing: if y is assumed to have ability for z, then
z is possible in a given world and y may be allowed, advised, etc. to z.
Poder has become lexicalized in the expressions puede que 'maybe' and
puede ser que 'it may be that', which can only convey EP, as illustrated in
examples (36) & (37).
(36) Pero ya eso la policía no te explica nada. Si alguna vez coincides
con algún amigo, amigo te hablo que lo has visto dos o tres veces -
entonces puede que a lo mejor te cuente algo. (M24,pl0)
'But the police don't explain anything to you. If you happen to run
into a friend, friend I mean someone you've seen two or three times
- then he may perhaps tell you something.'
(37) Pues aquí ese coche puede ser que sobre el setenta y dos al setenta
y cuatro como mucho dejaron de fabricarlo. (M9,A353)
'Well here, that car, maybe they stopped making it around seventy
two or seventy four at the latest.'
Support for the lexicalized status of puede que and puede ser que is
offered by the fact that they may only have third person singular morphology,
and that they are subjectless. With respect to negation, only puede ser que
may be negated and, as it is epistemic possibility, negation affects only p.
Examples (38-39) are illustrative.
(38) a. *No puede que Juan lo engañe.
b. Puede que Juan no lo engañe.
'It is possible that John is not cheating him'
(39) a. No puede ser que Juan lo engañe.
b. Puede ser que Juan no lo engañe.
a. 'It can't be possible that John is cheating him'
b. 'It is possible that John is not cheating him'

3.2 Conclusion

The data examined indicate that the most frequent CM of poder when used in
Present tense form is root possibility (58 of 110 cases studied). Ability, EP,
and mitigation account for 9%, 8%, and 4%, respectively. Furthermore, 9
Interpretation of 'poder' and 'deber' 85

examples are indeterminate between ability and RP, and 20 examples are
indeterminate between permission and RP.
Two further points must be emphasized: (1) Some of the 58 cases of RP
are in the periphery of this CM, close to readings of permission or ability; and
(2) The nature of the data (conversational discourse) must have consequences
on the frequency of occurrence of the CMs identified, such that other dis­
course genres would most likely yield different percentages of distribution of
CMs.
It is clear that we are dealing with intersecting contextual sets. On the
other hand, it is not clear to me that a strict prototype approach is the most
appropriate one to account for the facts. Indeed, a prototype model does not
acknowledge the relevance or validity of necessary or sufficient conditions in
the interpretation of CMs. Nevertheless, there do appear to be certain neces­
sary contextual features in every CM, and there are cases when the necessary
conditions are also sufficient to identify a CM as, for instance, 'permission': a
specific strong deontic source, a specific agent, direct speech, a dynamic
situation, positive consequences of p (ex. 40); or 'epistemic possibility':
non-agentive, non-intentional 'doer', stative situation, negative consequences
of p (ex. 41).

(40) [Teacher to child]

Puedes salir de la sala ahora.
'You may leave the room now'
(41) El mundo puede estar al borde de una crisis nuclear.
'The world may be about to suffer a nuclear disaster'
What happens to be the case is that these sets of contextual conditions are
frequently not a question of plus or minus in conversation, but of more or less.
It seems that the contextual features which are necessary conditions in the
permission and ability sets —deontic source and evidentiality— are precisely
the fuzzy ones. The fuzziness of contexts plus the logical relations between
ability, necessity, mitigation, permission, RP and EP often lead to intersecting
CMs.

4. Deber

Like must in English, deber can be used in either an epistemic or a deontic

sense, as shown in examples (42-43). An example of the type of (42) involves
86 Carmen Silva-Corvalán

'epistemic necessity', while an example of the type of (41) (with poder)

involves 'epistemic possibility' (cf. Lyons 1977: 791). However, given that
necessity and possibility are related notions, I have chosen to use the term
epistemic possibility in the semantic analysis of deber as well.
(42) [Loud car engine noises]
Ese debe ser Fernando. Epistemic possibility
'That must be Fernando.'
Interpretation'.
On the basis of frequent previous associations between loud engine
noises and Fernando's car, speaker confidently infers that the
person in the loud car is Fernando.
(43) Debes estudiar para aprobar el examen. Deontic obligation/necessity
'You must study to pass the exam.'
Interpretation'.
Common knowledge indicates that it is required/necessary to study
in order to pass an exam.
The IM of deber has some points of overlap with that of poder, but while
poder 'does not preclude X', deber 'favors X', 'requires X', or 'entails X'.
That is, while poder communicates the speaker's lack of strong belief in the
possibility of actualization of p, deber communicates confidence in the reali­
zation of p; it has connotations of 'very likely', 'necessary', and 'appropriate'.
The contribution of deber to the meaning of modalized p appears to be, then,
to require X as essential in the light of a set of circumstances C, interpreted
under a system of laws K, as represented in the formula: K (C requires X).
This definition is slightly different from the one proposed by Perkins
(1982:255) for the core meaning of must: K (C entails X). I have substituted
the term 'requires' for 'entails' in order to avoid misunderstanding due to a
reading of 'entails' in its strict logical sense. I use 'requires' in the sense of
'demanding as necessary or essential'. Thus, deber presents the IM 'requires'
and it relates actualization of X to a set of circumstances C, such that with
reference to a set of principles K, the circumstances C require X. As in the
case of poder, the values of K and C are recognized in context; these variables
(which include morphosyntactic, semantic and prosodic features) plus the
modal chosen determine whether X is interpreted under a dynamic, deontic or
epistemic modality.
Example (44) illustrates how the definition of deber would lead to an
interpretation of epistemic possibility assuming certain values for K and C.
Interpretation of 'poder' and 'deber' 87

(44) Tener un coche allí [en USA] debe ser algo alucinante. (M25,p5)
'To have a car there [in the USA] must be out of this world.'
IM: K (C requires X)
(i) K: rational laws (capacity to infer from available evidence)
(ii) C: (a) absence of deontic source
(b) preceding discourse gives some indication that p is true
(c) y = inanimate subject; z = stative; time orientation: generic
(iii) X: having a car in the USA is out of this world
CM: given humans' rational capacity to infer the likelihood of p
from available evidence, I interpret that the speaker intends to
convey the message that y ('to have a car in the USA') is very
probably z (' be out of this world'). (Epistemic possibility
sense).
By postulating for deber a core meaning of 'requirement of X', it is
possible to account for such CMs as obligation, advice, and probability,
depending on whether a number of contextual circumstances prompt a read­
ing of more or less forcible requirement. These circumstances include: (1)
Animacy and agentivity of y; (2) Nature of the deontic source, if any; (3)
Speaker approving or disapproving attitude towards p; (4) Adverse or positive
consequences of p.

4.1 Deber in Present tense form

Deber occurs infrequently in the data examined. Epistemic possibility seems

to be more frequently expressed by the lexicalized expression puede que,
while the auxiliary tener que 'to have to' is more commonly used to convey
deontic necessity. Only 26 cases of Present tense affirmative deber (de), plus
6 negative examples are attested in the data. The CMs of the 26 positive
examples range from 'high likelihood' (24 examples) to 'obligation/neces­
sity' (2 examples). The negative examples range from 'obligation/necessity'
to 'advice.'

4.1.1 Epistemic possibility to necessity

Spanish grammars note a difference between deber 'must' (obligation) and
deber de 'must' (possibility), but at the same time acknowledge that this
lexical opposition is being lost. The data examined support this observation
(cf. Sirbu 1988). The opposition deber - deber de is disappearing in the
88 Carmen Silva-Corvalán

Present tense, but appears to be more stable in other tenses (see Section 5 on
tense and modality). Deber de occurs in only 4 of the 24 examples of
epistemic possibility (see example 45). Furthermore, two examples with
deber de in the Imperfect Indicative do not convey possibility but rather
advice or obligation, as shown in (46).
(45) A: Es que se para [el coche] y luego-
B: xxx (incomprehensible)
A: Sí. Es que debe de estar estropeado el freno. (M24,p27)
A: 'It's that it stalls [the car] and then-'
B: xxx (unintelligible)
A: 'Yes. It's that the brake must be broken.'
(46) Yo exactamente no lo sé. Tendría que ser [el referendum] a prime­
ros del año que viene o algo así. Debían de solucionarlo antes de
terminar la jefatura de ellos [los socialistas] .(M24,p29)
'I don't know exactly. It [the referendum] would have to be at the
beginning of next year or something like that. They [the socialists]
should resolve this question before the end of their term in office.'
Deber is more strictly modal in its semantics than poder in that it is in all
contexts non-factual, i.e., propositions modalized by deber cannot be inter­
preted as 'actualized' ,15 This semantic difference accounts for the fact that one
of deber's CMs is EP, but not RP, which is related to ability and evidentiality.
In its EP contexts, deber may be preceded by negation but is not affected by it.
Examples (47) a-c are illustrative (see note 11).
(47) a. El alto debe ser Pepe.
b. El alto no debe ser Pepe.
c. *El alto debe no ser Pepe.
a. T h e tall one must be Pepe.'
b. 'The tall one can't be Pepe.'
c. *'The tall one must not be Pepe.'
The core contextual features associated with epistemic possibility are as
follows:
(i) K: rational laws
(ii) C: (a) evidence that p may be true
(b) absence of deontic source
Interpretation of 'poder' and 'deber' 89

(c) speaker attitude towards p is neutral

(d) adverse consequences of p
(e) y = inanimate; z = stative; time orientation: non-past
A human subject, a non-stative situation, speaker disapproval of not-p,
and adverse consequences of not-p determine readings which move away
from assessment of epistemic possibility towards the necessity end of the CM
continuum, examples (48a) and (48b) illustrate such a case.
(48) A: a. Porque se estropean las construcciones de agua [en La
Granja]. Por lo visto ya están muy pasadas de agua y pues
se deben estropear. Entonces las cuidan.
B: Las están arreglando.
A: b. Exactamente, las deben cuidar. (M24,p34)
A: a. 'Because the fountains break [at La Granja]. It's obvious
that they're too old (oversaturated) and so, of course they
{must) break. So they take care of them.'
B: 'They're fixing them.'
A: b. 'Exactly, they have to (must) take care of them.'
In (48a), a combination of core factors: inanimate subject, speaker neu­
tral attitude (i.e. neither approval nor disapproval) towards p, adverse conse­
quences of p, and evidence favoring the truth of p, determine a CM of
epistemic possibility. In (48b), on the other hand, C (human subject, non
-stative situation, adverse consequences of not-p) allows an interpretation of
necessity.

4.1.2 Mitigation
Changes in the contextual circumstances of deber in (48b): animate and
agentive subject, a possible though weak deontic source, strong evidence that
p is true, adverse consequences of not-p, bring about a CM which is indeter­
minate between high likelihood, categorical necessity or requirement, and
mitigation.
Note that example (48) illustrates intriguing uses of deber in utterances
which repeat the content of an immediately preceding non-modalized p. This
appears to indicate that in certain cases deber is used to convey a degree of
speaker uncertainty in the truth of p as a pragmatic strategy for mitigating the
degree of assertiveness of the utterance. The communicative purposes which
motivate the use of this strategy need to be further investigated. The choice of
90 Carmen Silva-Corvalán

deber over poder, which may also be interpreted to convey 'mitigation',

appears to be related to the higher degree of certainty implied by the IM of
deber.

4.1.3 Obligation to advice

The six negative constructions with deber in the data all express weak
obligation or advice:
(49) Nadie debe pensar por ti, ni nadie debe decidir por ti nunca, en
ninguna circunstancia. (M18,p27)
'No one should think for you, and no one should decide for you
ever, under no circumstances.'
I will not say much here about the expression of deontic obligation. I
would like to point out, however, that haber que 'to be necessary' and tener
que 'to have to' typically express necessity and obligation in affirmative
constructions in Present tense, as in (50) (cf. Sirbu 1988).
(50) Por ejemplo, hay que dar tres pagas al año, que dice el gobierno,
pues tres pagas al año se tienen que dar. Que hay que dar un mes
de permiso, pues un mes de permiso se tiene que dar. (M9,pl7)
'For instance, it is necessary to give three extra paychecks every
year, the government says, well then three extra paychecks have to
be given. And it's necessary to give one month of leave, well then
one month of leave has to be given.'
What is relevant, however, is that tener que also occurs, albeit quite
infrequently, in possibility contexts, as seen in (51):
(51) Y me pasó. Me pegó una pasada un 600, que yo digo, 'Pero la
madre del cordero, ese tío tiene que llevar un turbo ahí detrás. '
(M9,p3)
'And he passed me. A 600 passed me so fast, that I go, "Good
heavens, that guy must have a turbo behind".'
There are several indications that the use of tener que in possibility
contexts may be a recent development in Spanish: (a) The fact that neither
tener que nor the main verb Infinitive can be negated (exx. 52a-b); (b)
Infrequent occurrence of tener que in contexts of possibility; and (c) This
'meaning' of possibility is not acknowledged in current grammars and dic­
tionaries.16
Interpretation of 'poder' and 'deber7 91

(52) a. *Ese tío va tan lento que no tiene que sentirse muy bien.
b. *Ese tío va tan lento que tiene que no sentirse muy bien.
a. 'That guy's going so slowly that he doesn't have to be feeling
well'
b. 'That guy's going so slowly that he has to not be feeling well'

4.2 Summary

There appear to be a number of contextual sets of circumstances associated

with different CMs of deber. It seems possible to identify core contextual
circumstances invariably associated with epistemic possibility. By contrast,
fuzzy contexts make obligation, advice, and mitigation not always easy to
differentiate. As in the case oí poder, agentivity, evidentiality, and the possi­
ble consequences of p play an important part in the establishment of CMs.
Furthermore, ex. 46 suggests how verb morphology interacts with lexical
modalities. The next section explores this question further.

5. Modality and tense mood aspect

This section discusses the co-occurrence of poder and deber with verbal
markers of tense, mood and aspect other than 'Present tense'. Proposing IMs
for these verbal affixes is beyond the scope of this study. Therefore, my
analysis will consider only what appears to be their contextual contribution.

5.1 Non-finite forms and Imperative

I have observed in the introductory section that EP and RP are not possible
CMs when the modals are in Infinitive form (ex. 53). This restriction extends
to the Present Participle (ex. 54). Furthermore, neither modal is interpretable
in the Imperative (ex. 55). Thus, one might say that, like English modals,
poder and deber are 'defective' in this respect: they lack CMs of possibility in
the Infinitive and the Present Participle, and lack the Imperative altogether. In
the Infinitive (53) and the Present Participle (54) only root modality CMs are
inferred (e.g. obligation, ability). With regard to the Past Participle, the
behavior of poder and deber differs: while deber appears to retain many of its
possible CMs, including EP (56), poder in the Past Participle may convey
ability (57), but cannot be interpreted to convey RP (58) nor permission.
92 Carmen Silva-Corvalán

(53) Juan quiere deber/poder hacerlo.

'John wants to have the obligation/be able to do it.'
(54) Juan está debiendo/pudiendo hacerlo hace tiempo.
'John is having to/being able to do it for a long time now.'
(55) *Debe/puede hacerlo!.
*'Have to (must)/be able to (can) do it!'
(56) El árbol ha debido caerse con la lluvia.
lit: the tree has had to fall because of the rain
'The tree must have fallen because of the rain.'
(57) Juan ha podido irse.
'John has been able to leave.'
(58) *El árbol ha podido caerse con la lluvia.
* T h e tree has had the ability to fall because of the rain.'
Not a possible CM: The tree must have fallen because of the rain.
Verb morphology has been shown to be modal in that it can contribute to
the proposition the meaning of 'more or less assertiveness', where assertive­
ness is defined as speaker belief or confidence in the truth of the proposition.
Degrees of assertiveness are pragmatically inferred to convey degrees of
hypotheticality. These two notions correlate with verb morphology roughly as
in the scale below (cf. Fleischman 1989; Klein-Andreu 1986; Silva-Corvalán
1989).
Greatest assertiveness
factual Preterite Indicative
Present indicative
least hypothetical Future indicative
Imperfect indicative
Present subjunctive
most hypothetical Conditional Indicative
Imperfect subjunctive
Least assertiveness
In addition, verb morphology expresses tense and aspect oppositions,
roughly as follows: past time perspective (Preterite, Imperfect), omni-temporal
or atemporal time perspective (Present), future time perspective (Future, Condi­
tional, and perhaps also the Subjunctive forms). In regard to morphological
aspect, it may be argued that Preterite and Future are perfective,17 Imperfect
and Conditional Indicative, and Subjunctive forms are Imperfective, and Present
Indicative seems to be unmarked for aspect (cf. Silva-Corvalán 1991).
Interpretation of 'poder' and 'deber' 93

Let us now examine examples of poder and deber in non-Present form.

5.2 Imperfect and Preterite Indicative

31 examples of poder in Imperfect Indicative were examined: 6 are non-past

time oriented and exhibit desiderative (ex. 59) and RP CMs (cf. Bybee, this
volume); the rest have past time orientation and convey CMs of ability,
permission, RP (ex. 60), and obligation or advisability (this last one only in
one negative example, 61).

(59) Nos podíamos parecer a Argentina y, nosotros, a Estados Unidos

en eso, en eso, ¡en el dinero! (M9,A446)
'I wish we were (lit: we could be) like Argentine and, we, like the
United States, in that, in that, with respect to money!'
(60) Y estaba - todo la ca-, las calles estaban heladas. Y teníamos que
dar saltos, porque nos escurríamos y nos podíamos matar, ¿ com­
prendes? (M10,p6)
'And it was - the whole st- the streets were icy. And we had to jump
around, because it was slippery and we could kill ourselves, you
see?'
(61) Y cuando llegué a Madrid, ya definitivamente pues me planteé el
problema que yo no podía estar mirando al techo. Tenía que hacer
algo, ¿no? (M10,p3)
'And when I returned to Madrid, I definitely faced the question that
I couldn't just stare at the ceiling. I had to do something, right?'
Example (59) is interesting in that it has a CM not identified for poder in
Present tense, namely desiderative, in contexts with future time orientation.
This CM is inferred when the speaker's attitude is clearly approving of p. If
speaker attitude is neutral, however, the CM of podía with future time
orientation is simply possibility, as in (62).
(62) Si estás aburrido podías ir al cine.
'If you're bored you could go to the movies.'
Although it appears as if in Spanish root possibility in the past is marked
by the Imperfect morpheme (-bal-ía), this is not so. Past time orientation is
inferred from other signals in the context, as illustrated by the minimal pair
sentences in (63), where (63 a) has a future time orientation (signalled by el
próximo año 'next year'), while (63 b) may only be past (signalled by antes
'before, in the past').
94 Carmen Silva-Corvalán

(63) a. Si te apoyaba la ITT, podías ser Presidente el próximo año.

'If ITT supported you, you could be President next year.'
b. Si antes te apoyaba la ITT, podías ser Presidente, pero ahora
ya ese apoyo no sirve.
'If in the past ITT supported you, you could be President, but
now their support is useless.'
Halliday (1970:337) notes that "modality itself is not subject to variation
in tense". Contrary to Halliday, Coates (1983:107) discusses, for instance,
could as expressing, among other meanings, root possibility in the past, i.e., as
'past' of can. Example (64) is from Coates (ex. ii, p. 108).
(64) Past of can = Root Possibility
With all but one of the cookers the grid of the grill pan could be at
one of two possible distances from the heat. The exception was the
Cannon, which had four available positions.
'Exceptuando una de las cocinas, la rejilla del grill podía estar en
una de dos posibles distancias del calor. La excepción era la
Cannon, que tenía cuatro posiciones posibles.'
It seems to me that it is more appropriate to state that could is compatible
with past time contexts, just as podía is in Spanish, since past time in (64) is
indeed indicated by the verb forms was and had in the second sentence.
In Spanish, however, poder may occur with Preterite morphology and
unambiguously refer to past root possibility in certain contexts which do not
contain any other time reference (cf. Bybee's (this volume) interesting study
of the development of the hypothetical senses of past modals in English).
Compare examples (65) & (66) (fuller contexts are given in examples 69-70).
(65) Pudo marcharse.
could-3psg leave
'He was able to leave'
(66) Pudo pasar algo terrible.
might happen something terrible
'Something terrible might've happened.'
Example (65) conveys factuality of the event. By contrast, contextual
features, namely inanimate subject, stative verb, adverse consequences of p,
and speaker disapproval of p, determine a root possibility interpretation of
Interpretation of 'poder' and 'deber' 95

example (66), which is counter-factual; the Preterite suffix indicates that this
is root possibility in the past.
I have proposed that the IM of poder implies that a certain amount of
effort must be expended in 'not precluding X'. The relation between this IM
and the CM of (66) is, then, easy to establish: speakers do not assume that
entities may spend effort in the actualization of an event whose consequences
are adverse, so the only possible reading for (66) is of non-actualization or
non-factuality. Thus, this CM does not appear to be fuzzy, but rather a
category with clear boundaries defined by three features: adverse conse­
quences of p, speaker's disapproval of p, Preterite morphology. This analysis
is confirmed by the acceptability of (67) and (68 a), and the unacceptability
(under 'normal circumstances') of (68 b).

(67) Con esfuerzo, pudo marcharse finalmente.

'After trying hard, he was finally able to leave.'
(68) a. Estoy contenta, porque pudo pasar algo terrible, pero no pasó.
'I'm happy because something terrible might've happened, but
it didn't.'
b. * Estoy contenta porque pudo pasar algo terrible.
*'I'm happy because something terrible could happen'
I have excerpted the modalized propositions in (65-66) from actual
occurrences in the corpus (exx. 69 & 70) in order to contrast them in minimal
contexts, thus highlighting the contribution of adverse vs. neutral conse­
quences of p.
(69) Tuvo esa suerte y se pudo marchar. Le dieron un lectorado. (M10,p3)
'He was lucky and managed to leave. They gave him a lectureship.'
(70) Fue muy grave aquello. Pudo pasar algo terrible, porque si los
militares se echan a la calle - con seguridad ocurría una masacre.
Yo no sé qué hubiera ocurrido. (M10,pl3)
'That was really a critical situation. Something terrible could've
happened, because if the military get onto the streets - surely there
would've been a massacre. I don't know what would've hap­
pened.'
Adverse consequences of p is a crucial feature for CM root possibility
with Preterite poder. Despite the animacy of the subject, while example (71)
allows the inference that the situation did occur, (72) can only convey possi-
 96 Carmen Silva-Corvalán

bility (unless, of course, the speaker had wanted to become extremely poor,
but I would not consider this to be a 'normal social circumstance').
(71) Pudimos quedarnos en Inglaterra.
'We could stay in England.'
(72) Pudimos quedarnos en la miseria. [La situación estaba muy mala.]
could-we stay-us in the poverty
'We could've gone broke.' [The situation was bad.]
In its actual, more extended context of occurrence, example (71) in fact
conveys root possibility: staying in England was possible but not actualized.
However, this has to be made explicit by the speaker (pero él no quiso 'but he
didn't want to'), as shown in the actual utterance (73).
(73) Pero pudimos quedarnos en Inglaterra, porque le ofrecieron una
permanencia, pero él no quiso. (M10,p4)
'But we could have stayed (lit: could stay) in England, because
they offered him a permanent position, but he didn't want to.'
Otherwise, possibility with Preterite poder may be expressed if the modal
is combined with the Perfect Infinitive, as in English, such that example (74)
does not then require any further context to be interpreted as a possibility
which existed in the past, i.e. as counterfactual in the present.18
(74) Pudimos habernos quedado.
'We could have stayed.'
There are no examples with Preterite poder plus Perfect Infinitive in the
data from Madrid. I note, however, that in my variety of Chilean Spanish, this
tense combination is required in examples (66), (70), (72), & (73). My
intuitions in this respect are confirmed by the responses of four native speak­
ers from Chile and Argentina to an elicitation test given to determine the
meanings associated with poder and deber in various tense forms. This may
reflect a difference between Madrid and Latin American Spanish, but conver­
sational data from this latter variety must be examined before any further
proposals can be made in this respect.
The Perfect Infinitive does occur, however, with Present, Imperfect, and
Conditional forms of poder. All examples have past time orientation and
stress the non-factuality or hypotheticality of p. This higher degree of hypotheti-
cality becomes more evident through a comparison of examples (60) and (75).
Interpretation of 'poder' and 'deber' 97

(75) Tenía yo esto aquí que se podía haber utilizado.

T had this here that could've been used.'
Imperfective aspect in example (60) allows the interpretation that 'killing
ourselves' is omnitemporally possible if the contextual circumstances re­
ferred to in the example obtain. By contrast, Perfective aspect of modalized p
in (75) conveys that the possibility of p existed but is no longer available.
Examples with deber in Preterite and Imperfect are scarce (19 cases). It is
marked for tense only when combined with ser 'to be' in contexts which
explicitly indicate uncertainty about p (ex. 76).
(76) Debió ser eso de las 10 u 11 de la noche [cuando empezaron los
dolores del parto]. Y, no sé si nació a las 8 de la mañana.
(M24,pl8)
'It must've been at around ten or eleven at night [when labor pains
started]. And, I think she was born at eight in the morning.'
All other cases of Preterite and Imperfect deber convey obligation or
advice, even when combined with Perfect Infinitive, as shown in (77). Other­
wise, they are lexically differentiated (i.e. by using deber de) in contexts of
possibility, as illustrated by (78) and (79).19
(77) Por eso a mí me parece que debiste haber grabado a personas muy
jóvenes -porque tienen el lenguaje cheli.(M10,pl5)
'That's why it seems to me that you should've recorded very young
people - because they speak this cheli language.'
(78) Y mi madre debió de nacer en el año once o doce, porque tiene 75
o 74 años. (M10,p2)
'And my mother must have been born in 1911 or 1912, because
she's 75 or 74 years old.'
(79) Y la cogieron en el aeropuerto justo al llegar a Las Palmas.
Entonces allí debía, debía de haberlo llevado [la droga]. No sé
cómo funciona esto. (M24,p9)
'And they caught her at the airport just when she was arriving at
Las Palmas. So she must have taken it [the drug] there. I don't
know how that works.'
98 Carmen Silva-Corvalán

In regard to the Preterite/Imperfect opposition, I can only suggest at this

point that they reflect two different degrees of speaker assertiveness: stronger
and weaker, respectively. Thus, mutatis mutandis, in the case of deber,
Preterite morphology turns the IM requirement into obligation, while Imper­
fect morphology makes it less assertive and allows a peripheral reading of
weak obligation.
In my variety of Spanish, where opposition with deber de has disap­
peared, the expression of epistemic possibility in the past retains deber in the
Present and marks tense with the Perfect Infinitive. Thus, examples (78) &
(79) are rendered as in (80) & (81), respectively.

(80) Debe haber sido eso de las 10 u 11 de la noche ...

must-Pres. have been at around ...
(81) Y mi madre debe haber nacido en el...
must-Pres. have been born in ...

5.3 Conditional and Future

There are no occurrences of deber with future morphology in the data, and
only one of poder, with CM permission:
(82) Entonces esa persona que quiera abortar - podrá abortar siempre
que esto lo diga en las primeras semanas de su embarazo. (M10,p7)
'So a person who may want to have an abortion - will be able to
have one provided she requests it during the first weeks of preg­
nancy.'

In example (82), permission is associated with future time orientation

and identification of a deontic source (the law). If time orientation is present,
Future morphology independently conveys possibility, i.e. when combined
with non-modal verbs (see note 17), and it does so with poder as well.
Consider examples (83) a-b.
(83) a. Si tú lo dices, será (be-Fut.) cierto.
'If you say so, it must be true.'
b. Si tú lo dices, podrá ser cierto.
poder-Fut. be true
'If you say so, it may be true.'
Interpretation of 'poder' and 'deber' 99

The translations provided for examples (83) a-b are intended to capture
the fact that poder, because it incorporates the notion of effort, makes p more
hypothetical than when it is modalized simply by Future morphology.
By contrast, deber in the Future does not appear to be compatible with an
epistemic modality context, as I show in (84). Only root modality CMs may
be inferred for deber with Future morphology (85).
(84) a. ?Si tú lo dices, deberá ser cierto.
b. Si tú lo dices, será cierto.
'If you say so, it must-fut be true.'
(85) La inauguración deberá ser mañana.
'The inauguration will have to be tomorrow'
There are 6 examples with poder in Conditional form; they all occur in
root possibility contexts. Indeed, since Conditional morphology conveys very
weak assertiveness, it is not surprising that in every case it communicates
probability of p, regardless of the context. That is, non-modal verbs modal­
ized by Conditional morphology are interpreted to refer to hypothetical events
as well. Example (86) illustrates.
(86) Podrían hacerme una virguería - se suele decir una virguería, o
sea una cosa bonita -, arreglarme los dientes, quedarían perfectos.
(Mll,p27)
'They could do me a 'virguería' - they usually say 'virguería', you
know, something pretty -, fix my teeth, they would look perfect.'
On the other hand, deber in the Conditional may receive a deontic or an
epistemic interpretation depending on the same contextual features estab­
lished for Present deber. The contribution of Conditional morphology is that
of adding weak speaker assertiveness which, in a deontic context, creates a
CM of mitigated obligation or advice (ex. 87), and in an epistemic context, a
CM of weaker confidence in the truth of p (ex. 88).20
(87) El gobierno debería dar una explicación. (M5,p21)
'The government should give an expla