Volume 8 Issue 1 (2010)
Note: Linguistic Discovery uses Unicode characters
to represent phonetic symbols. Please see Optimizing Display
for requirements to accurately reproduce this page.
Semantic Maps and the Identification of Cross-Linguistic
Evidentiality and its Relation to Epistemic Modality
Cross-linguistic generic categories like evidentiality, tense,
aspect, number, and person are entrenched in linguistic theory. However, it is
not clear whether there is much empirical substance to them. There is a
remarkable lack of criteria for what counts as a category. This paper tries to
show that semantic maps can be used to give empirical substance to claims about
cross-linguistic generic categories. It is argued that, as falsifiable
cross-linguistic generalizations, semantic maps provide us with a criterion for
categorial status and category membership and also provide us with a basis for
identifying relations between different categories. However, it is also argued
that there are limits to the use of semantic maps in evaluating claims about
cross-linguistic generic categories, and that the criterion for categorial status
and category membership provided by semantic maps ultimately needs to be
supplemented by other criteria. In its argumentation, this paper focuses on the
category of evidentiality and on the relation between evidentiality and
In the most comprehensive study of evidentiality so far,
Aikhenvald (2004) as one of her central points claims that "evidentiality is a
category in its own right […]" (Aikhenvald 2004:7). Aikhenvald discusses a
large amount of data that may be taken intuitively to support her claim. She
also refers to studies that argue specifically in favor of setting boundaries
between evidentiality and epistemic modality (de Haan 1999) or between
evidentiality and mirativity (Lazard 1999, 2001, DeLancey 2001). But in fact,
she does not present a single explicit argument in support of her claim. She
does not give a single explicit criterion for what it means to be “a
category in its own right” and is therefore unable to demonstrate that any
criterion is met.
This is by no means meant as a critique of Aikhenvald alone. In general,
discussions of cross-linguistic categories like evidentiality, tense, aspect,
aktionsart, modality, number, and person are devoid of any interest in making
precise what it is that qualifies them to be considered categories in the first
place (cf. Boye and Harder 2009:15). The categories are normally taken to be
notionally coherent. That is, each category is taken to cover a range of
meanings that are similar in that they can be defined in terms of the same
abstract notion— what Bache (1997:108) calls a “category
concept”. For instance, evidentiality is typically taken to cover meanings
that can be defined in terms of the notion “source of information”
(e.g. Willett 1988, Bybee et al. 1994, Aikhenvald 2004). But if similarity in
meaning were all there was to categorial status, claims such as that made by
Aikhenvald would be trivial. Aikhenvald’s claim itself would be nothing
but a statement that there is a range of meanings that can be defined in terms
of the notion of source of
When we talk about
cross-linguistic categories like evidentiality, tense, aspect, etc., we imply
that there is some adequate generalization over language-specific facts which
goes beyond similarity in meaning. We imply that the notional coherence of a
range of meanings is linguistically significant in some way.
Following Whorf (1956:113), I shall refer to cross-linguistic categories
like evidentiality, tense, and aspect as cross-linguistic
categories. In what follows, I argue that semantic
maps can be used to give substance to claims about such categories. More
precisely, I argue that, as falsifiable cross-linguistic generalizations, they
may provide us with an empirically based criterion for categorial status and
category membership as well as for identifying relations between distinct
categories. In my argumentation, I focus on Aikhenvald’s aforementioned
claim about evidentiality. In particular, I try to show how a semantic map of
epistemic expressions can provide substantial input into the ongoing debate on
the relation between evidentiality and epistemic modality (van der Auwera and
Plungian 1998, de Haan 1999, Dendale and Tasmowski 2001b; Guentchéva and
In Section 2, I specify what I mean by the notion of
“cross-linguistic generic category” and argue that the notion, or
any equivalent notion, cannot be dispensed with in linguistic theory. In Section
3, I propose what I shall refer to as “semantic-map continuity” as
one criterion for identifying cross-linguistic generic categories. In Section 4, subsequently,
I present a semantic map of epistemic expressions. And in Section 5, I show how
the criterion and the map can be used to evaluate claims about evidentiality and
the relation between evidentiality and epistemic modality. In Section 6, I go on
to discuss the limits to semantic-map continuity as a criterion for identifying
cross-linguistic generic categories, and eventually I propose an additional
criterion. Finally, Section 7 is a brief conclusion.
2. Cross-Linguistic Generic
The term “category” serves a number of different
purposes in linguistic theory. For instance, the term is used both to designate
specific linguistic “values” like past, present, and future—
Whorf’s (1956:113) "specific" categories—and to designate
“groups of related values” like tense or time—Whorf’s
it is used both in
descriptions of language-specific structures and in cross-linguistic
generalizations; it is used both for linguistic phenomena and for the cognitive
and communicative phenomena that are assumed to motivate them; it is used both
to talk about purely theoretical constructs and to talk about entities that are
assumed to have an ontological reality. And it is far from always clear in which
way the term is used.
In the present paper, the term “cross-linguistic generic
category” is restricted to entities like evidentiality, tense, and aspect.
It is taken to refer to cross-linguistic generalizations over distinct
language-specific expressions that are related within as well as across the
languages in which they are found. In so far as cross-linguistic generic
categories are based on relations between expressions across languages, they
must take the form of meaning generalizations—abstract meaning categories;
arguably, linguistic expressions can be related across languages only in terms
of their meanings—as opposed to their morphosyntactic
But in so far as they
are also based on relations between expressions within specific languages,
cross-linguistic generic categories are endowed with a claim of being
significant for the description of language-specific phenomena that do not
necessarily pertain to meaning only. The assumption that cross-linguistic
generalizations reveal something about—and may thus be accounted for in
terms of—cognitive and communicative structures is well-founded, in so far
as human language is understood as a cognitive and social phenomenon. However,
this does not entail that cross-linguistic generalizations can be
straightforwardly interpreted as descriptions of cognitive and social
structures, as is often done. In the present paper, cross-linguistic generic
categories are therefore conceived of as pertaining to linguistic phenomena
only, as opposed to cognitive or social phenomena. And as generalizations over
language-specific facts, they
are conceived of entirely as the linguist’s theoretical construct rather
than as corresponding to any entity which is assumed to have an ontological
Not all linguists would subscribe to the view that there is a need for
cross-linguistic generic categories in linguistic theory. Among others, Bybee
and her collaborators have argued that categories like evidentiality, tense, and
aspect can easily be dispensed with in language-specific and therefore also
cross-linguistic description. The argument turns on the traditional association
of those categories with morphosyntactically delimited systems of linguistic
expressions. While such systems are often found, the argument goes, they are
usually not notionally coherent (Bybee 1985:17), and, whether or not they are
notionally coherent, they must be considered “epiphenomenal”:
“the clues to understanding the logic of grammar are to be found in the
rich particulars of form and meaning and the dynamics of their
coevolution” (Bybee et al. 1994:1-2). For Bybee and her collaborators,
linguistic universals are found only at the level of values like past, present,
and future, as opposed to the level of groups of related values like tense
(Bybee et al. 1994:3). They do not reject that values fall into distinct,
notionally coherent groups. In fact, they readily group the universal values
they study, “grams”, into notionally coherent groups: “aspects
and tenses”, “agent-oriented modalities”, “moods”
and “evidentials” (Bybee et al. 1994:316-324). What they reject is
the usefulness of any generalization over distinct linguistic values which, like
Aikhenvald’s category of evidentiality, is intended to go beyond
registration of notional coherence.
However, notional generalizations are in themselves essentially
intuition-based and subjective. Bybee et al.’s generalization over meaning
paraphrases such as “direct evidence” and “indirect
evidence” in terms of the notion of source of information may intuitively
seem perfectly sound. But if the notional generalization itself were all we were
left with, there would be no rational way to judge whether the generalization is
adequate or not. If it were impossible to come up with empirical constraints on
notional generalizations, we would have to abandon as unscientific the use of
terms such as “evidentiality”, “tense”, and
“aspect” from linguistics. Bybee et al. would probably agree that if
these terms were deleted from their book, it would have fatal consequences for
its readability. Luckily, it is possible to come up with empirical constraints.
Below I give one reason why cross-linguistic generic categories, in the sense
specified above, cannot be dispensed with in linguistic theory.
The reason has to do with the fact that, as acknowledged by Bybee and
her collaborators, some language-specific systems are actually notionally
coherent. Whether it is an epiphenomenon or not, the fact that notionally
related linguistic expressions occasionally, or perhaps even often, cluster in
morphosyntactically delimited systems does not follow trivially from their
notional relation. Consider, for instance, the often-cited Ngiyambaa
(Australian) clitics -
gara ‘sensory evidence’ and
Dhan ‘spoken, or by extension written evidence’, illustrated
in (1) and (2) respectively.
Ngiyambaa (Australian; Donaldson
‘One can see you were
‘You are said to have been
These two clitics make up a morphosyntactic system which is
delimited by two properties: 1) The clitics mutually exclude each other
(Donaldson 1980:244); 2) They always follow so‑called “knowledge
clitics” (Donaldson 1980:241) when they co-occur with
This system is notionally
coherent in so far as both clitics have meanings that can be described in terms
of the notion of source of information. What is important here is that systems
like this, that are both morphosyntactically and notionally coherent, need not
exist. Theoretically, all expressions of source of information in the languages
of the world might either be completely absent in morphosyntactic systems or be
found in systems that are not notionally coherent. For instance, one might
perfectly well imagine a situation where there were no morphosyntactic
properties that delimit the two clitics mentioned above from the Ngiyambaa
“knowledge clitics”. In support of this, it is far from always being
the case that the morphosyntactic properties which delimit a notionally coherent
system can be straightforwardly accounted for as a consequence of the notional
feature in terms of which the members of the system are related. The second of
the two properties that delimit Ngiyambaa -
gara and -
other Ngiyambaa clitics cannot. The fact that the two clitics are mutually
exclusive can hardly be seen as a consequence of the fact that both clitics
express source of information. In many languages, two or more expressions of
source of information may co-occur. This is the case in Eastern Pomo (Hokan),
where “the hearsay or reportative evidential” suffix
-·le can follow one of two other evidential suffixes, including
“the inferential evidential” suffix -
(i)ne, as in (3)
Eastern Pomo (Hokan; McLendon
‘He must have simply left his
daughter-in-law there, they say’.
In Qiang (Sino-Tibetan), likewise, the marker of
“inferential evidence” and the marker of “visual
evidence” can be used together, as in (4) (LaPolla 2003:69; cf. Aikhenvald
2004:87-96, and Boye 2006:180-196 for discussion and additional
Qiang (Sino-Tibetan; LaPolla
‘Oh, he WAS playing a
To repeat, the
existence of language-specific systems like the one found in Ngiyambaa that are
both morphosyntactically and notionally coherent is a nontrivial fact. As a
nontrivial fact it ought not to be neglected. Rather, systems like that found in
Ngiyambaa suggest that a notional generalization over linguistic expressions of
source of information is not only intuitively sound, but also linguistically
It is the implication of linguistic significance which makes claims
about cross-linguistic generic categories different from claims about notional
coherence and which thus makes cross-linguistic generic categories indispensable
in linguistic theory.
3. Semantic-Map Continuity as a
Criterion for Categorial Status and Category Membership
In the following I outline how semantic maps can be used to
give empirical substance to the notion of cross-linguistic generic
categories—how they can be employed in an empirically based criterion for
categorial status and category membership.
The criterion is this:
Criterion of semantic-map continuity for status as and membership in a
The meanings comprised by a cross-linguistic
category together make up a continuous region of a semantic
That is to say, any claim about cross-linguistic categorial
status would be rejected if in a semantic map at least one of the hypothesized
members were not immediately adjacent – i.e. linked by what Haspelmath
(2003) calls “connecting lines”—to any other member. Consider,
for instance, Haspelmath’s semantic map of indefinite pronouns in Figure 1
(Haspelmath 1997: 119).
Figure 1. Haspelmath’s (1997) semantic map of
If I were to claim the significance of identifying a
cross-linguistic generic (sub)category of indefinite pronouns which comprises
the meaning “specific known” and the meaning “irrealis
non-specific”, but not the meaning “specific unknown”, my
claim could be shot down immediately by checking it with the criterion above
relative to Haspelmath’s map: “specific known” and
“irrealis non-specific” do not together make up a continuous region
of the map. If, on the other hand, I took the category to comprise also the
meaning “specific unknown”, my claim would meet the criterion in so
far as the three meanings “specific known”, “specific
unknown”, and “irrealis non-specific” together make up a
continuous region of Haspelmath’s map.
The reason why semantic-map continuity may serve as a criterion for
categorial status and category membership is that, just like the notionally
coherent morphosyntactic systems discussed in Section 2, it is a linguistic
empirical finding that does not follow, in any trivial way at least, from
notional coherence. For instance, the fact that in Haspelmath’s semantic
map of indefinite pronouns the meanings “specific known”,
“specific unknown”, and “irrealis non-specific” together
make up a continuous region does not follow from the (apparent) appropriateness
of generalizing over the three meanings in terms of the notion of
(non-)specificity. The structure of a semantic map is an empirical result
obtained by first identifying and generalizing over a number of comparable
meanings across languages and then studying which meanings are directly related
to each other in terms of synchronic polyfunctionality or diachronic change. A
cross-linguistic notional generalization is the result of a different and far
more subjective operation. Like the operation of constructing a semantic map,
this operation involves an initial round of identifying and generalizing over a
number of comparable meanings across languages. Subsequently, however, it
submits the output of the first, already quite subjective, round to a second
subjective round of generalizing over the distinct meaning generalizations.
There is no reason to think that the result of constructing a notional
generalization can be used to predict the result of constructing a semantic
The criterion of semantic-map continuity thus provides us with a
guarantee of empirical linguistic significance which purely notional
generalizations over linguistic expressions cannot provide themselves. It
provides a basis for reinforcing claims about notional coherence into claims
linguistically significant notional coherence—that is,
into claims about cross-linguistic generic categories.
To my knowledge, semantic maps have been employed in such reinforcing of
claims only once and only implicitly. In their paper on “Modality’s
semantic map”, van der Auwera and Plungian (1998) define modality in terms
of the notions of possibility and necessity (van der Auwera and Plungian
1998:80). They argue that meanings that can be described in terms of the former
notion and meanings that can be described in terms of the latter notion are
adjacent in a semantic map: diachronically, they argue, expressions of necessity
meaning may change into expressions of possibility meaning, and vice versa, and
synchronically, expressions are found that are polyfunctional with respect to
the two meanings (van der Auwera and Plungian 1998:97-104). In effect, then,
they argue that modality defined in terms of the notions of possibility and
necessity meets the criterion of semantic-map continuity for being identified as
a cross-linguistic generic category.
4. A semantic map of epistemic
I shall now return to Aikhenvald’s claim about
evidentiality as a category in its own right. In Section 5 I show how the
criterion of semantic-map continuity and a relevant semantic map can be used to
evaluate this claim in general and claims about the relation between
evidentiality and epistemic modality in particular. To this end, below I present
a relevant semantic map: a map of epistemic expressions.
Epistemic expressions are taken to comprise linguistic items and
constructions (grammatical as well as lexical) that express either source of
information or degree of (un)certainty, or both. Thus, they comprise the English
expressions emphasized in (5)-(10).
English expressions of source of
The former president of France
have been bald.
I hear that the former president of
France was bald.
The former president of France was
English expressions of degree of
The former president of France
probable that the former president
of France was bald.
The former president of France was
The semantic map of epistemic expressions below is based on
data from 52 languages representing 35 phyla (according to the classification in
However, the map is
also compatible with data from a great number of languages discussed in
Givón (1982), Akatsuka (1985), Bybee et al. (1994), and Aikhenvald
(2004). The map may be represented as in Figure 2.
Figure 2. A semantic map of epistemic
In Figure 2, as in Figure 1, the numbered lines represent
what Haspelmath (2003) refers to as “connecting lines”. The labels
between the lines characterize epistemic meanings, or meaning regions. The map
is a large-scale map. As indicated in the parentheses added to some of the
labels, more fine-graded distinctions can be drawn for at least three of the
meaning regions distinguished (cf. Anderson 1986 for a detailed semantic map of
evidential expressions). However, this, as well as the ultimate empirical
tenability of the map, is irrelevant for the present illustrative
In the rest of this section, focusing on the evidential meaning regions
(direct evidence and indirect evidence), I will exemplify the linguistic basis
for the distinctions (Section 4.1) and the connecting lines (Section 4.2) found
in the map.
4.1 Distinctions made in the
semantic map of epistemic expressions
All distinctions made in the semantic map of epistemic
expressions are linguistically significant. Below, I will exemplify the
linguistic basis for drawing a distinction between direct evidence and indirect
The distinction is taken to correspond to Willett’s (1988:57)
distinction between “direct evidence” and “indirect
evidence”, and roughly to Aikhenvald’s (2004:65) distinction between
“direct” or “firsthand” evidence on the one hand and
“non-firsthand” evidence on the other. It is based on the finding
that cross-linguistically, linguistic expressions are often found that indicate
either 1) (possibly a subtype of) direct evidence, but not (any subtype of)
indirect evidence, or 2) (possibly a subtype of) indirect evidence, but not (any
subtype of) direct
An example of the first type of expression is the Ngiyambaa clitic
gara ‘sensory evidence’, illustrated in (1) above. Other
examples are found in Imbabura Quechua (Quechuan), where both of the clitics
-má ‘emphatic first-hand information’ and
‘first-hand information’ indicate direct evidence (Cole
1982:163-166) and in Qiang (Sino-Tibetan), where direct evidence is indicated by
the verbal suffix
‘visual’ (LaPolla 2003). In still other languages direct
evidence is marked by the absence of other epistemic markers. In Hixkaryana
(Carib), for instance, “zero marking”, outside constructions
involving the “nonpast uncertainty suffix”, “marks
‘eyewitness’ in contrast to ‘hearsay’ ”
(Derbyshire 1979:143). In Turkish (Altaic), direct evidence is marked by
“Ø” outside past-tense contexts (e.g. Kornfilt 1997:377-378).
And “any declarative utterance in Bella Coola [Salishan] implies that the
speaker has witnessed what he reports” (Davis and Saunders 1975:15; cf.
Aikhenvald 2004:70-79 for a general discussion of “zero realization”
in evidential systems).
An example of the second type of expression is the Ngiyambaa clitic
Dhan ‘spoken, or by extension written evidence’, illustrated
in (2) above. Other examples are the Hixkaryana particles
‘deduction’ (Derbyshire 1979:143-145), the Basque particles
omen ‘citationnel’ and
(Jendraschek 2003:29, Oyharçabal 2003:316), and the West Greenlandic
-sima ‘apparently’ (Fortescue
2003:292-294). These clitics, particles, and suffixes all indicate a subtype of
indirect evidence—inferential or reportive evidence. Expressions such as
the Mangarayi (Australian) “past irrealis” construction (Merlan
1982:136-151) and the Montagnais (Algic) suffix -
shapan seem to indicate
indirect evidence in general, that is, to cover both the meanings of inferential
and reportive evidence (cf. James et al. 2001:238 on the use of -
in the Montaignais of Betsiamites and Lower North Shore).
4.2 Connecting lines found in
the semantic map of epistemic expressions
Every connecting line found in the semantic map of epistemic
expressions has been established on the grounds that a number of
language-specific expressions are polyfunctional with respect to, or move
diachronically between, the meanings connected by the
Below, I exemplify the
linguistic basis for identifying the connecting line between direct evidence and
indirect evidence (connecting line 1 in Figure 2).
For this connecting line I have found no unequivocal diachronic
evidence. That is, I have found no expressions for which diachronic movement
from the meaning of direct evidence to the meaning of indirect, or vice versa,
can be claimed unequivocally. However, I have found plenty of expressions that
are polyfunctional with respect to the two meanings, and it may be speculated
that at least some of these expressions are the result of diachronic extension
of an originally more narrow evidential meaning.
Examples of polyfunctional expressions include the Sherpa (Sino-Tibetan)
+nok, which is used to indicate direct evidence in the
present tense but indirect evidence in the past (Woodbury 1986:190-196), and the
Kashaya (Hokan) suffix
-qă, which may be used to mark “an
inference based on circumstances or evidence found apart, in space and time,
from the actual state or event”, but may also be used as “a default
category for [direct] evidence through senses other than those that have a
specific suffix (Visual and Auditory)” (Oswalt 1986:38). In Lega
(Niger-Congo), likewise, the particle
ampó among other meanings
covers both “direct sensory evidence, particularly from sight or
hearing”, as in (11), and “strong inferential [and thus indirect]
evidence” (Botne 1997:517), as in (12).
Lega (Niger-Congo; Botne
‘She’s assuredly pounding rice (I
can hear it)’.
‘(It's evident, as his odd behavior
indicates, (that)) Moke forgot that Amisi went to Pangi’.
A number of other examples of expressions that cover both
the meanings of direct and indirect evidence are found in what Aikhenvald
(2004:29-31) calls evidential “A2” systems. These systems of
grammatical expressions of source of information consist of two members which
differ in terms of a contrast between direct-visual evidence on the one hand and indirect
evidence and direct-non-visual evidence on the other hand (cf. Table 2.1 in
Aikhenvald 2004:65). Thus, the “non-visual” member cuts across the
distinction between direct and indirect evidence.
5. Evidentiality and the
semantic map of epistemic expressions
Armed with the, as yet unfalsified, semantic map of
epistemic expressions and the criterion of semantic-map continuity, we are now
ready to confront Aikhenvald’s claim about evidentiality as a category in
its own right in general, and claims about the relation between evidentiality
and epistemic modality in particular.
5.1 Mapping evidentiality onto
the semantic map of epistemic expressions
As for Aikhenvald’s claim, it meets the criterion of
semantic-map continuity when it is evaluated against the proposed semantic map
of epistemic expressions.
When we apply the criterion of semantic-map continuity to a proposed
cross-linguistic generic category, we ask whether the meaning regions covered by
the proposed category together constitute a continuum in a relevant semantic
map. That is, we ask whether each of the meaning regions distinguished within
the category is connected in a relevant semantic map to at least one of the
other meaning regions distinguished. In the case of Aikhenvald’s category
of evidentiality, the answer to this question is yes. Aikhenvald defines
evidentiality in terms of two features: 1) the notion of source of information
and 2) grammatical status. Only the former feature is relevant here since, as
conceived of by Aikhenvald (and others), it includes the meaning regions of
direct evidence and indirect evidence as evidential, and excludes all other
meaning regions as non-evidential. In the semantic map of epistemic expressions
represented in Figure 2, these two meaning regions together make up a continuum.
They are connected to each other by connecting line 1. Thus, Aikhenvald’s
cross-linguistic category of evidentiality may be mapped onto the semantic map
of epistemic expressions, as illustrated in Figure 3.
Figure 3. Aikhenvald’s cross-linguistic category of
evidentiality mapped onto the semantic map of epistemic
Aikhenvald’s evidentiality meets the criterion of
semantic-map continuity then. However, this does not entail that her claim that
evidentiality is a category in its own right is hereby confirmed beyond dispute.
First, it might turn out that the semantic map of epistemic expressions can be
falsified when confronted with more data or with a reanalysis of the data on
which it is based. On a revised map the meaning of direct evidence and the
meaning of indirect evidence might not be connected by a connecting line.
Second, as will be demonstrated in Section 6, semantic-map continuity is not a
sufficient criterion for categorial status and category membership. Meeting the
criterion is not sufficient for being identified as a cross-linguistic generic
Still, semantic maps and the criterion of semantic map continuity can be
used to effectively reject some claims about cross-linguistic generic
5.2 Evaluating claims about
evidentiality by means of semantic maps
The categorial status of evidentiality has been intensively
discussed in connection with attempts to describe the relation between
evidentiality and epistemic modality. Several claims have been made about this
relation (cf. the overviews given in Dendale and Tasmowski 2001b:241-242, and
Nuyts 2006:10-12). Some works, like Aikhenvald’s, take evidentiality and
epistemic modality to be separate categories (e.g. de Haan 1999 and
2001:27, 35). Others take the former to include the latter (e.g. Papafragou
2000:121, and Ifantidou 2001:5-8), or the latter to include the former (e.g.
Palmer 1986:51, and Willett 1988:52). Still others take them to be separate but
overlapping categories (e.g. van der Auwera and Plungian 1998:85-86, and Palmer
2001:8-9). And a couple of works take evidentiality and epistemic modality to be
separate categories that are, however, both subcategories of the same
superordinate category (Hengeveld 1989, and Boye 2006). The variety of claims is
partly a result of the lack of empirically based criteria for identifying
cross-linguistic generic categories. Categories are identified and distinguished
mainly on the basis of notional coherence, and the choice between the claims is
thus to some degree a terminological choice between different opinions about
which notions—which ranges of meanings—the terms
‘evidentiality’ and ‘epistemic modality’ should be
Confronted with the criterion of semantic-map continuity and the
semantic map of epistemic expressions, at least one of the claims must be
rejected: van der Auwera and Plungian’s (1998) claim that evidentiality
and epistemic modality are separate but overlapping categories. Like
Aikhenvald’s category of evidentiality, van der Auwera and
Plungian’s categories of evidentiality and epistemic modality actually
turn out to meet the criterion of semantic-map continuity when evaluated against
the semantic map of epistemic expressions. As for the category of evidentiality,
van der Auwera and Plungian share Aikhenvald’s conception of it and
explicitly take it to comprise the meaning regions direct and indirect evidence
(as well as other meanings that are arguably covered by the two regions) (van
der Auwera and Plungian 1998:85). Thus, van der Auwera and Plungian’s
categor of evidentiality covers the same continuous region of the semantic map of epistemic
expressions as does Aikhenvald’s category (compare Figure 3 above with
Figure 4 below). As for the category of epistemic modality, van der Auwera and
Plungian define it as comprising two meaning regions: “epistemic
possibility” (which they take to amount to “uncertainty”) and
“epistemic necessity” (which they take to amount to
“[c]ertainty and a relatively high degree of probability”) (van der
Auwera and Plungian 1998:81). In so far as it comprises certainty, probability,
and epistemic possibility, then, this category also covers a continuous region
of the semantic map of epistemic expressions, as illustrated in Figure
Figure 4. Van der Auwera’s and Plungian’s (1998)
cross-linguistic categories of evidentiality and epistemic modality mapped onto
the semantic map of epistemic expressions
The reason for rejecting van der Auwera and Plungian’s
claim is that their idea of an overlap between evidentiality and epistemic
modality is incompatible with the semantic map of epistemic expressions.
According to van der Auwera and Plungian, the overlap is located in the
epistemic modal region of “epistemic necessity”—or
“[c]ertainty and a relatively high degree of probability”—and
the evidential region of inferential evidence, a subregion of indirect evidence.
These two regions, they claim, are identical—one amounts to the other (van
der Auwera and Plungian 1998:85-86; cf. Palmer 2001:8-9 on
“deductive” as an overlap region). However, in the semantic map of
epistemic expressions the two regions are clearly distinct, and the distinction
made between them is based on plenty of linguistic evidence. Expressions are
found that indicate degree of certainty but not source of information, and
expressions are found that indicate source of information but not degree of
certainty. As an example of the latter, the Kashaya inferential-evidence suffix
-qă, discussed in Section 4.2 above,
has no implications of
degree of certainty. What it implies is “merely lack of higher ranking
evidence” (Oswalt 1986:38; cf. de Haan 1999 and Aikhenvald 2004:153-193
for additional evidence against the overlap view).
6. The limits to semantic maps
in identifying cross-linguistic generic categories
Claims about cross-linguistic generic categories are claims
about linguistically significant meaning generalizations. Semantic-map
continuity is a nontrivial linguistic fact, and in the preceding section I have
tried to show how semantic maps and the criterion of semantic-map continuity can
be used to evaluate whether or not a given meaning generalization is in fact
However, semantic maps are not an unequivocal means for identifying
categories and relations between them. As already mentioned, the criterion of
semantic-map continuity is not a sufficient criterion for categorial status and
To see this, consider again Aikhenvald’s claim that evidentiality
is a category in its own right. As demonstrated in Section 5.1, this claim meets
the criterion of semantic-map continuity when evaluated against the proposed
semantic map of epistemic expressions. As noted in the same section, however,
this is no proof that her claim is correct in any sense. Claims that are
incompatible with Aikhenvald’s also meet the criterion. For instance, the
following claim does:
Aikhenvald’s category of evidentiality is not a category in
its own right. Rather, the meaning regions direct evidence and indirect evidence
are comprised by a cross-linguistic category of epistemicity which in addition
comprises the meaning regions certainty, probability, and epistemic
Arguably, the range of meanings covered by the category of
epistemicity proposed in this claim is notionally coherent—the meaning
regions direct evidence, indirect evidence, certainty, probability, and
epistemic possibility can all be defined in terms of a notion like the
philosophers’ “justificatory support”, for instance (cf. Boye
2006 for details). As illustrated in Figure 5, the category of epistemicity maps
onto a continuous region of the semantic map of epistemic expressions.
Figure 5. The
cross-linguistic category of epistemicity
mapped onto the semantic
map of epistemic expressions
The following claim
(made in Boye 2006) likewise meets the criterion of semantic-map
Aikhenvald’s category of evidentiality is not a category in
its own right—at least not in the sense that it is not comprised by a
superordinate category. Rather, a category of epistemicity comprises
Aikhenvald’s evidentiality as a subcategory side by side with a
subcategory of epistemic modality (defined as to comprise the meanings of
certainty, probability and epistemic possibility).
All of the three proposed categories arguably cover
notionally coherent ranges of meanings, and each of the categories maps onto a
continuous region of the semantic map of epistemic expressions, as illustrated
in Figure 6.
Figure 6. The cross-linguistic categories of evidentiality,
epistemic modality, and epistemicity mapped onto the semantic map of epistemic
What disqualifies semantic-map continuity as a sufficient
criterion for categorial status and category membership is that a one-to-one
correspondence between the range of meanings defined by notional coherence and
the range of meanings defined by semantic-map continuity cannot be expected (cf.
Section 3). First, the fact that two meaning regions are notionally
similar and together make up a continuous region of a semantic map does not
entail that a category covers
only the two regions at hand. For instance,
the epistemic modal meaning regions certainty and probability can both be
defined in terms of the notion of degree of certainty, and together they make up
a continuous region of the semantic map of epistemic expressions (cf. Figure 6),
but this does not entail that epistemic modality covers only these two regions
and not the region of epistemic possibility as well. Second, the fact
that two meaning regions together make up a continuous region of a semantic map
does not entail that they are notionally coherent. For instance, expressions of
indirect evidence may develop from, among other things, perfects, resultatives,
and past tenses (Aikhenvald 2004:279-281). Accordingly, any semantic map that
covered the meaning of indirect evidence as well as resultative meaning, perfect
meaning, and past tense meaning would have to include connecting lines between
the latter meanings and the former. But clearly, the meaning of indirect
evidence is notionally distinct from its three source meanings, and to my
knowledge, it has never been claimed to belong to the same category as any of
them. Semantic-map continuity may cut across category borders then. Connecting
lines are not necessarily category internal, but may be category
If we want to be able to qualify essentially subjective claims about
notional coherence into empirically based claims about cross-linguistic generic
categories, the criterion of semantic-map continuity will not do the job alone.
We need other criteria. An obvious candidate would draw on the fact discussed in
Section 2 that some language-specific systems are actually notionally coherent.
The criterion might be formulated something like this:
Morphosyntactic criterion for status as and
membership in a cross-linguistic category
Meanings comprised by a cross-linguistic
category are frequently found with all members of language-specific
morphosyntactically delimited systems of expressions.
As we need other criteria anyway, it is natural to ask
whether we need the criterion of semantic-map continuity in the first place. It
is not a sufficient criterion, but is it a necessary criterion for categorial
status and category membership? I suggest that it is. However, the answer is
ultimately dependent on an exact definition of cross-linguistic generic
categories. Cross-linguistic generic categories are taken here to refer to
linguistically significant meaning generalizations. The question as to whether
or not the semantic-map continuity criterion needs to be met in order for something to be
identified as a category or as a member of a category is a question of whether
or not “linguistic significance” is taken to necessarily include
semantic-map continuity. Semantic-map continuity is a nontrivial fact which can
be used to reinforce claims about notional coherence into claims about
notional significance, but it is not
the only such fact.
As noted in Section 1, there is a remarkable lack of
criteria for what counts as a cross-linguistic generic category, and categories
like evidentiality, tense, aspect, modality, person, and number tend to be
identified and distinguished from each other mainly on the basis of notional
coherence. One consequence of this is that discussions about such categories are
often hampered by lack of consensus as to what the categories cover (discussions
about modality are a notorious case in point). Notional generalizations are
essentially subjective, and every scholar is free to come up with his or her own
definition. Another consequence is that, as enlightening and intuitively sound
as they may be, claims about the functional or cognitive motivations for the
categories are fundamentally hollow. As long as we have no means for assessing
whether a notional generalization is linguistically significant or not, it might
seem a complete waste of time to try to motivate it or to draw conclusions about
communicative or cognitive phenomena on the basis of it. A third consequence,
finally, is that the notion of cross-linguistic generic categories itself
becomes exposed to suspicions that it is superfluous. Unless we come up with
criteria for distinguishing categories as linguistically significant notional
generalizations from purely notional generalizations, we cannot blame scholars
like Bybee for dismissing the former and sticking with the latter.
In this paper I have argued that cross-linguistic generic categories
cannot be dispensed with in linguistic theory, and that semantic maps can be
used to give substance to claims about them. Focusing on the category of
evidentiality and its relation to the category of epistemic modality, I have
tried to show that semantic maps may provide us with an empirically based
criterion for categorial status and category membership as well as with a
falsifiable basis for identifying relations between different categories.
However, I have also argued that semantic maps are not an unequivocal means for
identifying categories and relations between them, and that the criterion of
semantic-map continuity thus needs to be supplemented with additional
Cross-linguistic generic categories should not be conceived of as
pre-established in the sense of Haspelmath (2007). Rather, they are the result
of what Lazard (2001:365) calls “the only way out of the difficulty of
linguistic typology (especially grammatical typology)”. As outlined by
Lazard, this way starts with the formation of hypotheses and intuition-based
notional generalizations, and it goes on with empirical evaluation of these
hypotheses and generalizations until they are ultimately validated. Not only can
semantic maps help us pass the starting line. They can also take us a good deal
Abbreviations used in this paper and not standardized in the
Leipzig Glossing Rules: AG agent, EVID evidence,
INFEVID inferential evidence, PAT patient,
RECPST recent past, REPEVID reportative evidence,
SENSEVID sensory evidence, VISEVID visual evidence.
Aikhenvald, Alexandra Y. 2004. Evidentiality. Oxford: Oxford
Aikhenvald, Alexandra Y. and R.M.W. Dixon (eds.). 2003. Studies in
Evidentiality. Amsterdam: Benjamins.
Akatsuka, Noriko. 1985. Conditionals and the epistemic scale.
Language 61/3.625-639. doi:10.2307/414388
Anderson, Lloyd B. 1986. Evidentials, paths of change, and mental
maps: Typologically regular asymmetries. Chafe and Nichols 1986,
Bache, Carl. 1997. The study of aspect, tense and action: Towards a
theory of the semantics of grammatical categories. Frankfurt am Main:
Botne, Robert. 1997. Evidentiality and epistemic modality in Lega.
Studies in Language 21/3.509-532. doi:10.1075/sl.21.3.03bot
Boye, Kasper. 2006. Epistemic meaning: A cross-linguistic study.
Ph.D. thesis. Copenhagen: University of Copenhagen.
Boye, Kasper and Peter Harder. 2009. Evidentiality: Linguistic
categories and grammaticalization. Functions of Language
Bybee, Joan L. 1985. On the nature of grammatical categories.
Proceedings of the second eastern states conference on linguistics, ed. by
Soonja Choi, Dan Devitt, Wynn Janis, Terry McCoy and Zheng-sheng Zhang, 17-34.
Buffalo, NY: The State University of New York.
Bybee, Joan L., Revere D. Perkins and William Pagliuca. 1994. The
evolution of grammar: Tense, aspect and modality in the languages of the world.
Chicago: University of Chicago Press.
Chafe, Wallace and Johanna Nichols (eds.). 1986. Evidentiality: The
linguistic coding of epistemology. Norwood: Ablex.
Cole, Peter. 1982. Imbabura Quechua. Amsterdam:
Croft, William. 2003. Typology and universals, 2nd edition.
Cambridge: Cambridge University Press.
Davis, Philip and Ross Saunders. 1975. Bella Coola deictic usage.
Rice University Studies 61.13-35.
de Haan, Ferdinand. 1999. Evidentiality and Epistemic Modality:
Setting boundaries. Southwest Journal of Linguistics 18.83-101.
Dendale, Patrick. and Liliane Tasmowski (eds.). 2001a. On
Evidentiality. Journal of Pragmatics 33/3 (special issue). Amsterdam:
-----. 2001b. Introduction: Evidentiality and related notions.
Dendale and Tasmowski 2001a, 339-348.
DeLancey, Scott. 2001. The mirative and evidentiality. Dendale and
Tasmowski 2001a, 369‑382.
Derbyshire, Desmond. C. 1979. Hixkaryana. Amsterdam:
Donaldson, Tamsin. 1980. Ngiyambaa. Cambridge: Cambridge University
Fortescue, Michael. 2003. Evidentiality in West Greenlandic: A case
of scattered coding. Aikhenvald and Dixon 2003, 291-306.
Givón, Talmy. 1982. Evidentiality and epistemic space.
Studies in Language 6/1.23-49. doi:10.1075/sl.6.1.03giv
Gordon, Raymond G., Jr. (ed.). 2005. Ethnologue: Languages of the
world, 15th edition. Dallas, TX: SIL International. Online version:
Guentchéva, Zlatka (ed.) 1996. L'énonciation
médiatisée. Paris: Peeters.
Guentchéva, Zlatka and Jon Landaburu. (eds.) 2007a.
L'énonciation médiatisée, vol. 2: Le traitement
épistémologique de l'information: Illustrations
amérindiennes et caucasiennes. Louvain: Peeters.
-----. 2007b. Introduction. Guentchéva and Landaburu 2007a,
Haspelmath, Martin. 1997. Indefinite pronouns. Oxford: Clarendon
-----. 2003. The geometry of grammatical meaning: Semantic maps and
cross-linguistic comparison. The new psychology of language: Cognitive and
functional approaches to language structure, ed. by Michael Tomasello, vol. 2,
211-242. London: Erlbaum.
-----. 2007. Pre-established categories don't exist: Consequences
for language description and typology. Linguistic Typology
Heine, Bernd, Ulrike Claudi and Friederike Hünemeyer. 1991.
Grammaticalization: A conceptual framework. Chicago: University of Chicago
Hengeveld, Kees. 1989. Layers and operators in Functional Grammar.
Journal of Linguistics 25.127-157. doi:10.1017/s0022226700012123
Ifantidou, Elly. 2001. Evidentials and relevance. Amsterdam:
James, Deborah, Sandra Clarke and Marguerite MacKenzie. 2001. The
encoding of information source in Algonquian: Evidentials in
Cree/Montagnais/Naskapi. International Journal of American Linguistics
Jendraschek, Gerd. 2003. La modalité
épistémique en basque. Munich: Lincom.
Kornfilt, Jaklin. 1997. Turkish. London: Routledge.
LaPolla, Randy J. 2003. Evidentiality in Qiang. Aikhenvald and
Dixon 2003, 63-78.
Lazard, Gilbert. 1999. Mirativity, evidentiality, mediativity, or
other? Linguistic Typology 3/1.91-110. doi:10.1515/lity.19220.127.116.11
-----. 2001. On the grammaticalization of evidentiality. Dendale
and Tasmowski 2001a, 359-367.
McLendon, Sally. 2003. Evidentials in Eastern Pomo with a
comparative survey of the category in other Pomoan languages. Aikhenvald and
Dixon 2003, 101-130.
Merlan, Francesca. 1982. Mangarayi. Amsterdam:
Nuyts, Jan. 2001. Epistemic modality, language, and
conceptualization. Amsterdam: Benjamins.
-----. 2006. Modality: Overview and linguistic issues. The
expression of Modality, ed. by William Frawley, 1-26. Berlin: Mouton de
Oswalt, Robert L. 1986.
The evidential system of Kashaya. Chafe and Nichols 1986, 29-45.
Oyharçabal, Bernard. 2003. Tense, aspect and mood. A grammar
of Basque, ed. by José. I. Hualde and Jon Ortiz de Urbina, 249-284.
Berlin: Mouton de Gruyter.
Palmer, Frank R. 1986. Mood and modality. Cambridge: Cambridge
-----. 2001. Mood and modality, 2nd edition. Cambridge: Cambridge
Papafragou, Anna. 2000. Modality: Issues in the
semantics-pragmatics interface. Oxford: Elsevier.
Slobin, Dan. I. 1997. The origins of grammaticizable notions:
Beyond the individual mind. The crosslinguistic study of language acquisition,
ed. by Dan I. Slobin, vol. 5: Expanding the contexts, 265-323. London:
van der Auwera, Johan and Vladimir A. Plungian. 1998. Modality's
semantic map. Linguistic Typology 2/1.79-124. doi:10.1515/lity.1918.104.22.168
Whorf, Benjamin L. 1956a . Some verbal categories of Hopi.
Language, thought, and reality: Selected writings of Benjamin Lee Whorf, ed. by
John B. Carroll, 112-124. Chicago: MIT Press.
Willett, Thomas. 1988. A cross-linguistic survey of the
grammaticization of evidentiality. Studies in Language 12/1.51-97. doi:10.1075/sl.12.1.04wil
Woodbury, Anthony C. 1986. Interactions of tense and evidentiality:
A study of Sherpa and English. Chafe and Nichols 1986, 188-202.
Author's Contact Information:
considers evidentiality a grammatical category and thus takes it to cover only
meaning of grammatical expressions (for instance, affixes or auxiliaries)
(e.g. Aikhenvald 2004:11). But this has no impact on her claim that
evidentiality is a category in its own right. It is interesting, of course, that
the notion of “source of information” seems to provide a
generalization over grammatical expressions in a large number of languages (cf.
e.g. Chafe and Nichols 1986, Willett 1988, Guentchéva 1996, Dendale and
Tasmowski 2001a, Aikhenvald and Dixon 2003, Guentchéva and Landaburu
2007a). More generally, it is an interesting hypothesis that
cross-linguistically the meanings of grammatical elements can be described in
terms of a limited range of notions (Slobin's (1997) “grammaticizable
notions”; cf. e.g. Heine et al. 1991:32-39, Bybee, et al. 1994:10, Croft
2003:225). Surely, however, these observations and hypotheses are independent of
any of the normal implications of claims about categorial status.
his distinction between specific and generic categories as follows: "SPECIFIC
CATEGORY: an individual class of [...] types, e.g. passive voice, durative
aspect, vs. GENERIC CATEGORY: a higher hierarchy formed by grouping classes of
similar or complementary types, e.g. voice, aspect" (Whorf 1956:113). While the
distinction is a cornerstone in linguistic theory, it is rarely made explicit,
and Whorf's terminology is used even more rarely. In a recent paper, for
instance, Haspelmath (2007:123) refers to generic categories as
properties are purely structural phenomena whereas meanings are not. As argued
by the old structuralists and much more recently by functional typologists (cf.
e.g. Haspelmath 2007 for a recent statement), structure is language-specific
whereas substance is potentially universal.
Just like linguistic meanings, phonetic features can also be
compared across languages in terms of a common substance. Thus, linguistic
expressions can be related across languages also in terms of their phonetic
properties. However, phonetic relations are irrelevant for the identification of
cross-linguistic generic categories.
According to Donaldson
(1980), the Ngiyambaa “knowledge clitics” comprise
‘exclamative, what’ and -
ga ‘ignorative, I don't
know’. To judge from Donaldon’s glossings, these clitics are
notionally clearly distinct from the clitics illustrated in (1) and (2),
gara ‘sensory evidence’ and -
Dhan ‘spoken, or
by extension written evidence’.
The languages are
(with the classifications in Gordon 2005 added in parentheses): Acoma (Keres),
Basque (Basque), Bella Coola (Salishan), Berbice Dutch Creole (Creole), Carib
(Carib), Cairene Egyptian Colloquial Arabic (Afro-Asiatic), Danish
(Indo-European), English (Indo-European), Finnish (Uralic), French
(Indo-European), German (Indo-European), Gooniyandi (Australian), Hanis
(Penutian), Hdi (Afro-Asiatic), Hidatsa (Siouan), Hixkaryana (Carib), Hopi
(Uto-Aztecan), Imbabura Quechua (Quechuan), Jacaltec (Mayan), Jarawara (Arauan),
Japanese (Japanese), Kannada (Dravidian), Kashaya (Hokan), Khasi
(Austro-Asiatic), Kobon (Trans-New Guinea), Kolyma Yukaghir (Yukaghir), Korean
(Isolate), Koyra Chiini (Nilo-Saharan), Ladakhi (Sino-Tibetan), Lango
(Nilo-Saharan), Lega (Niger-Congo), Limbu (Sino-Tibetan), Mangarayi
(Australian), Mapuche (Araucanian), Montagnais (Algic), Ngiyambaa (Australian),
Nunggubuyu (Australian), Qiang (Sino-Tibetan), Rapanui (Austronesian),
Shipibo-Konibo (Panoan), Sherpa (Sino-Tibetan), Slave (Na-Dene), Southern
Nambiquara (Nambiquaran), St'át'imcets (Salishan), Supyire (Niger-Congo),
Tamil (Dravidian), Tidore (West Papuan), Tukang Besi (Austronesian), Turkish
(Altaic), Wari' (Chapacura-Wanham), Western Apache (Na-Dene), West Greenlandic
This practice entails
that if all attested expressions with the meaning of direct evidence covered
also the meaning of indirect evidence, and all attested expressions with the
meaning of indirect evidence covered also the meaning of direct evidence, the
two meanings could not be distinguished in the semantic map.
In order to reduce the
risk that a connecting line between two distinguishable meanings is established
on the basis of misinterpreted or erroneous data, it should in practice be based
on several attested cases of polyfunctionality or diachronic movement between
the meanings. In principle, however, a connecting line could be established
between two meanings even if only one case of polyfunctionality or diachronic
movement between the meanings were attested. Connecting lines are normally taken
to represent functional or conceptual relations between the meanings they
connect, and the existence of functional or conceptual relations must be
independent of the cross-linguistic frequency with which they are actually
exploited in the development of polyfunctional linguistic expressions. (Even the
strength or closeness of functional or conceptual relations may be independent
of the cross-linguistic frequency with which the relations are
exploited–at least, any natural link has as yet to be
The author's contributions to the present issue of Linguistic Discovery were made possible by a grant from The Carlsberg Foundation. The author wishes to thank Andrej Malchukov and an anonymous reviewer for comments on a first draft of the present paper.