Chapter 1 Evidentiality and Information Source

In: Evidentials and Modals
  • 1 James Cook UniversityLanguage and Culture Research Centre


Reference to information source may be accomplished with a variety of means, including verbs referring to reports, claims, or opinions, adverbs, parentheticals, prepositional phrases or particles. In about one quarter of the world’s languages, marking information source is obligatory. These languages have a grammatical category of evidentiality. Other languages have evidential extensions of non-evidential categories—such as conditional in French. These ‘evidentiality strategies’ share the evidential meanings and often give rise to grammatical evidentials. The term ‘evidential’ primarily relates to information source as a closed grammatical system whose use is obligatory. The term ‘information source’ relates to the corresponding conceptual category. Expressions related to information source are heterogeneous and versatile, and may allow more detailed specification of various degrees of assumption, inference, opinion than do grammatical evidential systems, and often reliability, and speaker’s evaluation of information. The paper focuses on various aspects of expressing information source across the world’s languages.

Every language has an array of ways of referring to information source.* This may be accomplished with verbs referring to reports, claims, or opinions, with adverbs, with parentheticals, prepositional phrases or with particles. In about one quarter of the world’s languages, marking information source is obligatory. These languages have a grammatical category of evidentiality. Other languages have evidential extensions of non-evidential categories—such as conditional in French, perfect in Georgian and participles in Lithuanian. Just like non-grammatical expressions of information source, evidential extensions of non-evidential categories (known as evidential strategies) share the evidential meanings and not infrequently give rise to grammatical evidentials.

The term ‘evidential’ primarily relates to information source as a closed grammatical system whose use is obligatory. The term ‘information source’ relates to the corresponding conceptual category. This is akin to the distinction between the category of ‘tense’, as grammaticalized location in time, and the concept of ‘time’. Expressions related to information source are heterogeneous and versatile. They include closed classes of particles and modal verbs, and an open-ended array of verbs of opinion and belief. The term ‘lexical evidentiality’ is misleading in that it obscures these vital differences.

Extra-grammatical ways of marking information source may allow more detailed specification of various degrees of assumption, inference, opinion than do grammatical evidential systems, and often reliability, and speaker’s evaluation of information.

1 Information Source and Evidentiality

Every language has a way of saying how one knows what one is talking about, and what one thinks about what one knows. But the ways in which the information source can be expressed vary. Languages differ not in what one can say, but in what one must say—as stated by Boas (1938), one of the founders of modern linguistics and of the study of grammatical expression of information source: ‘grammar […] determines those aspects of each experience that must be expressed’ (Boas 1938:132). One language may have a two-term gender system, while another has five genders and a third makes no gender distinctions at all in its grammar—though ‘man’ and ‘woman’, and males and females are clearly distinguished in other ways. Along similar lines, some languages have grammatical tense, and others do not. But in any language, one can talk about time (see § 5, and Aikhenvald 2012: 44–45).

In at least one quarter of the world’s languages, marking a limited selection of information sources is obligatory. In Tariana (Arawak), Matses (Panoan), Makah (Wakashan), Hup (Makú), Quechua and Aymara, a clause without a marker of information source would not be acceptable to a native speaker. This is grammatical evidentiality—a brief summary is in § 2.

Verbal categories whose main meanings do not reflect information source can acquire evidential extensions. Giacalone Ramat & Topadze (2007) show how perfect aspect in Georgian regularly extends to cover evidential meaning associated with inference. This is an example of an evidential strategy fairly wide-spread in many languages of the Caucasus—see the summary in § 3.

Every language—no matter whether it has a fully grammaticalized evidential system, or evidential extensions of other categories—has an array of further ways of describing how one knows things.1

Terminological clarity is essential in any branch of linguistics: the importance of distinguishing information source and grammatical evidentials is highlighted in § 5. § 6 is a brief summary.

2 Evidentiality as a Closed Grammatical System: A Bird’s Eye View

As a category in its own right, evidentiality is a relatively recent ‘arrival’ on the linguistic scene—in contrast to other categories such as person, gender, number and tense which have been household concepts in linguistics for thousands of years (see, for instance, Robins 1967). This may well be the reason why the proper limits of evidentiality are still debated by some.

The idea of obligatory marking of information source goes back to Boas, and his sketch of Kwakiutl (1911: 443; 496). ‘The source, or nature, of human knowledge (known by actual experience, by hearsay, and by inference)’ was listed by Sapir (1921:108–109) alongside other grammatical concepts, such as person, modality, number and tense. Since Boas’s work, the notion of grammatical evidentiality has made its way into many grammars of North American Indian languages. Evidentiality is a recurrent feature of South American languages (detailed discussion is in Aikhenvald 2012: 248–278). It is also prominent in Turkic, Uralic, and Tibeto-Burman languages. Evidentiality systems in the languages of Africa have only been described relatively recently (see Botne, this volume; Storch and Coly 2014; Storch 2018). But for grammarians of European languages it remained largely unknown.

The term ‘evidential’ as a label for the grammatical category of information source was first introduced by Jakobson in 1957; and became established by the mid-60s (see Jacobsen 1986:4–7; Aikhenvald 2004: 10–17). Lazard (1957) was among the first French linguists to have discussed evidential meanings (‘inférenciel’), based on the material from Tajik, an Iranian language.

In languages with obligatory evidentiality, a closed set of information sources has to be marked in every clause—otherwise the clause is ungrammatical, or the speaker incompetent, or even not quite right in his mind (Weber 1986: 142). Evidentiality is a category in its own right, and not a subtype of modality, tense, or mood.2

Languages with grammatical evidentials divide into a number of types depending on how many information sources are assigned a distinct grammatical marking. Semantic parameters employed in languages with grammatical evidentiality cover physical senses, several types of inference and of report. The recurrent terms in the systems are:

  1. Visual covers information, or evidence, acquired through seeing.

  2. Sensory covers information obtained through hearing, and is typically extended to smell and taste, and sometimes also touch.

  3. Inference based on visible or tangible evidence or result.

  4. Assumption based on evidence other than visible results: this may include logical reasoning, assumption or simply general knowledge.

  5. Reported, for reported information with no reference to who it was reported by.

  6. Quotative, for reported information with an overt reference to the quoted source.

The maximum number of evidential terms in a grammatical system appears to be five. Recent studies in grammatical evidential systems have revealed the existence of further terms. For instance, Yongning Na (Mosuo), a Tibeto-Burman language (Lidz 2007), has a special term in its evidentiality system which covers exclusively ‘general knowledge’. Mamaindê, a Nambiquara language from Southern Amazonia, also has a special evidential for ‘general knowledge’. This language expresses secondhand and thirdhand information by using different morphological markers (Eberhard 2009, 2018).

Amazonian languages may have further terms. In the Southern Nambiquara dialect complex, there is an obligatory marking on the verb for, among others (Lowe 1999):

  • whether a statement is eyewitness—that is, implying that the speaker had seen the action they are reporting,

  • whether a statement is inferred or assumed, whereby ‘the speaker’s claim … is based either on seeing an associated simultaneous action and making an interpretation therefrom, or on seeing a set of circumstances which must have resulted from a previous action and making an inference; different suffixes mark these two options’,

  • whether it is reported, that is if ‘the speaker is simply passing on information they have heard from another speaker’, or

  • whether there is ‘internal support’—if ‘the speaker reports their ‘gut feeling’ that which they assert must be so’ (Lowe 1999: 275–276).

No spoken language has a special evidential to cover smell, taste, or feeling (not so in sign languages: Catalan sign language is reported to have a special evidential marking smell: Sherman Wilcox, p.c.). Evidentiality in sign languages and how it may be different from modality is a matter for further investigation.

Semantic parameters group together in various ways, depending on the system. The most straightforward grouping is found in three-term systems—where sensory parameters (I and II), inference (III and IV) and reported (V and VI) are grouped together in Quechua, Shilluk, Bora (Aikhenvald 2004: 145–146; 159–166). Numerous languages of Eurasia group parameters (IIVI) under a catch-all non-firsthand evidential, as, for example, do Abkhaz or Yukaghir.

Evidentials may not have epistemic extensions, to do with probability and speaker’s evaluation of the trustworthiness of information.3 This is frequently the case in smaller systems: for instance, the reported evidential in Estonian has strong connotations of unreliable information. Visual evidential in Quechua can refer to information the speaker vouches for. Not so in Tariana or Tucano: here visual evidential is to do with access to what one had seen rather than to certainty. A plethora of markers express doubt, certainty or lack thereof.

Reported evidential in Estonian has an overtone of doubt. This is akin to how ‘they say’ in English may imply that the speaker does not really believe what is being reported, or to how the ubiquitous dizque has overtones of doubt in many varieties of South American Spanish (Kany 1944: 171; Travis 2006; Babel 2009; and Olbertz 2005, 2007). In contrast, in Quechua, Shipibo-Konibo and Tariana, the reported evidential does not have any such overtones. As Valenzuela (2003: 57) remarks for Shipibo-Konibo, the selection of reported evidential over the direct evidential ‘does not indicate uncertainty or a lesser degree of reliability but simply reported information’. These languages have a plethora of other categories which express doubt, belief, disbelief and so on. See Chapter 5 of Aikhenvald (2004) for a survey of epistemic extensions, or lack thereof, in the grammatical evidentials in the world’s languages (further information about the types of systems and their meanings can be found in Aikhenvald 2018).

Just like most other grammatical categories, evidentials interrelate with mood (that is, clause types: Lyons 1977). The maximum number of evidential specifications tends to be distinguished in declarative main clauses. The most frequent evidential in commands is reported (‘do what someone else told you to’) (see Aikhenvald 2010: 138–141).

Future and non-indicative modalities—conditional, dubitative and so on—(not to be confused with moods) may allow fewer evidential specifications than the indicative. The maximum number of evidential specifications is found in past tenses.

When used with a first person subject, the non-visual, non-firsthand evidentials and reported evidentials in systems of varied types may acquire additional meanings to do with lack of intention, control, awareness and volition on the part of the speaker. Verbs covering internal states may require obligatory evidential choice depending on person.

For example, in Tariana (an Arawak language from north-west Amazonia) and adjacent East Tucanoan languages a visual evidential is not appropriate when talking about one’s internal states and feelings. If my tooth is hurting, I have to use a nonvisual evidential:






‘My tooth is hurting’ (nonvisual)

If I am talking about someone else’s toothache, I cannot use the nonvisual marker: I cannot feel the pain someone else is feeling. If I see someone suffering from toothache, I could say (2), using a visual evidential:






‘His tooth is hurting’ (visual: I can see that he is suffering from toothache)

If the person is gone to see a dentist, I may well use the inferred evidential to talk about their toothache: this is based on my inference that a trip to a dentist can only be the result of a toothache.












‘His tooth is hurting (inferred), he has already gone to a doctor’4

In examples (1)–(3), the choice of evidential correlates with person. That is, evidentials can be considered exponents of person, and have an implicit value of person markers.

Evidentials typically have sentential, or clausal, scope. Dependent clauses usually cannot have an evidential value different from that of a main clause. Having an NP within the scope of a grammatical evidential is highly unusual. One of the rare well-documented examples known so far comes from Jarawara (an Arawá language from South America: Dixon 2003; Aikhenvald 2004: 88; see also Aikhenvald 2014: 16–19 and Storch and Coly 2014 on the expression of non-propositional evidentiality).5

3 Evidential Strategies

Meanings to do with how people know things may be expressed without developing a dedicated form whose primary meaning is information source. Non-evidential categories frequently acquire evidential extensions. A conditional mood, or a perfect, or a passive can develop an evidential-like meaning as a ‘side effect’ (see the discussion in Lazard 1999). This kind of extension of a category is referred to as a ‘strategy’ (along similar lines, a language can have complementation strategies, relativisation strategies, imperative strategies, and so on).

One of the best-known examples is the conditional in French (also known as ‘conditionnel d’ information incertaine’) used to relate information obtained from another source for which the speaker does not take any responsibility (see Dendale 1993, and Dendale and Van Bogaert 2007). Conditional in Italian may also extend to cover reported information (see Squartini 2007).

Perfect aspect is a way of expressing non-firsthand evidential meanings in Georgian (Giacalone Ramat & Topadze 2007, Hewitt 1995: 259; 93). This development is shared with many Iranian, Turkic, and Northeast Caucasian languages (also see Comrie 1976:110; Aikhenvald 2004: 289–296).

Or the choice of a complementizer or a type of complement clause may serve to express meanings related to how one knows a particular fact. In English, different complement clauses distinguish an auditory and a hearsay meaning of the verb hear: saying I heard John cross the street implies that I did hear John stamping his feet, while I heard that John crossed the street implies a verbal report of the result. That is, a that-clause with perception verbs can refer only to indirect knowledge (see a concise analysis of complement clauses with verbs of perception in English in the context of complementation in general, by Dixon 2005: 270–271).6

Nominalizations and participles often develop connotations similar to non-firsthand evidentials. In his discussion of Lithuanian, Wiemer (2007) mentions participles ‘as semi-grammaticalized means of indicating hearsay or inferential meanings’. This is consistent with Gronemeyer’s (1997) and Timberlake’s (1982) analysis of Lithuanian passive participles. According to Mathiassen (1996: 134–135) and Gronemeyer (1997) active participles as head of predicate in Lithuanian have been reinterpreted as reported evidentials.

Marking of assertion and speaker’s authority correlates with speaker’s attitudes to information and—indirectly—to its sources. As shown by Pusch (2007), enunciative particles in Gascony Occitan mark speaker’s assertion intertwined with meanings related to expression of the ways in which information was acquired. ‘Assertivity’ in Gascony Occitan is not an evidential system; in Pusch’s words, ‘it oscillates between some kind of modality […] and evidentiality proper’.

Evidentiality strategies typically develop a range of meanings characteristic of reported and non-firsthand evidentials: they combine reference to inference and to verbal report. And they are not averse to having epistemic extensions to do with probability, and also expressing speaker’s attitudes to the veracity of what is being said.

Non-indicative moods and modalities, and also perfects, resultatives, passives and nominalizations tend to express inference based on results and on assumption. They may extend to cover hearsay, or secondhand, information.

Markers of reported speech, particles derived from ‘say’, and de-subordinated speech complements including nominalizations express primarily hearsay or secondhand information. They may extend to cover further meanings to do with inference based on results, and also assumption.

The range of meanings of the two groups of evidentiality strategies is summarized in Figure 1.1.


Figure 1.1

The semantic range of evidentiality strategies

No language has been found to have a special evidentiality strategy for each of the evidential meanings which can be expressed (IVI in § 2). Most of the features outlined for grammatical evidentials in § 2 are not characteristic of evidential strategies. They qualify as ‘incipient’ evidentials-in-the-making which tend to grammaticalize into a closed system of evidentials.

Over time, an evidential overtone of a non-evidential category may conventionalize as its major meaning. In other words, evidential strategies may develop into grammatical evidentials—as was the case in Lithuanian (Gronemeyer 1997). A future tense can give rise to a dedicated non-firsthand evidential, as happened in Abkhaz (Chirikba 2003: 262–264). And the enunciative particle que in Gascony Occitan (Pusch 2007) has the potential of developing into an evidential marker.

Not every extension to do with information source is an evidentiality strategy. Pietrandrea (2007) shows that the ‘epistemic future’ in Italian is not really an evidentiality strategy: its connections with information source are tenuous, and its evidential extensions always depend on the pragmatic context.

4 Further Ways of Expressing Information Source

Every language can express doubt, inference, and assumption. The means vary—from open classes of verbs, adverbs and adjectives (§ 4.1), and parentheticals (§ 4.2) to more restricted subsets of modal verbs and grammaticalized particles (§ 4.3). Speech report constructions in their varied guises are another, almost universal, device (§ 4.4) for talking about what one learnt from someone else.

4.1 Open Lexical Classes: Verbs, Adverbs, and Adjectives

Most languages have a large number of verbs expressing meanings linked to information source. Romance and Germanic languages are particularly rich in these—a multiplicity of such expressions in French involve penser ‘think’, trouver ‘think, judge’, avoir l’ impression ‘have the impression’ (also Dendale & Van Bogaert 2007, Pietrandrea 2007 and Giacalone Ramat & Topadze 2007).

English also has an immense array of reporting verbs and opinion verbs like think, suppose, find, claim, state or allege, in addition to verbs to do with seeming or appearing. Each of these is semantically versatile; they vary in their subtle grammatical properties (see Dixon 2005: 202–206). One can say It looks like rain, or This idea sounds good, or I hear you are getting married—each of these ways of saying things in English can be replicated in German, Dutch, French, Italian, Spanish and Swedish. These are the ways in which familiar European languages allow us to express some of the meanings which must be expressed grammatically in languages like Quechua, Tariana, Qiang, Western Apache and Shipibo-Konibo (where they form an obligatory closed system).

This does not imply that these ‘exotic’ languages are bereft of verbs of opinion, thinking, reasoning, claiming and so on. They are not—on the contrary, Tariana (Aikhenvald 2003) has a vast array of verbs to do with mental states. And one can use them to complement the restricted number of choices imposed by obligatory evidentials. Consider (4):






‘She (assumed) has already died, as I intuitively feel’ (my gut feeling tells me that she is dead)

By (4), I specify the fact that the use of assumed evidential is based on my intuitive suspicion (and not on a general assumption). By saying (5), I stress that the assumption—encoded in the evidential—is based on logical reasoning.





‘She (assumed) has already died, as I reason’ (that she is dead is a logical conclusion based on my reasoning)

Tariana has no evidential to describe intuition and reasoning. The lexical ways of marking information source are much more versatile than the grammatical options. The interaction between these two is what makes Tariana discourse fascinating (similar techniques are available in other languages—see papers in Aikhenvald & Dixon 2003). And this is also what makes the study of verbs expressing information source in French, by Dendale & Van Bogaert (2007), an enticing read. There are many more options in the details one may want to express though lexical means than through grammar.

Adverbial expressions in Italian express possibility, probability, doubt, and can also extend to refer to inference, assumption, validity of information and attitude to it—that is, they may be used to refer to information source (see Pietrandrea 2007). English adverbs reportedly, supposedly and allegedly and Estonian kuuldavasti ‘reportedly’ are cases in point. One can also opt to use an adjective to express a similar meaning: one hears reference to an alleged drug-dealer, or a supposedly false statement. The choices are many.

Prepositional constructions may express opinion, belief, inference and so on: compare Italian secondo me ‘according to me’, Portuguese ao meu ver (lit. to my seeing) ‘in my opinion’, and noun phrases involving prepositions anot and pasak in Lithuanian (see Wiemer 2007). These are arguably more epistemic than inferential evidential—but this is a matter of approach.

4.2 Parentheticals

European languages tend to have a plethora of parentheticals, such as English I think, I suppose, Spanish parece, Italian sembra and French dit-on and paraît-il ‘it appears’. A parenthetical—defined as ‘a word, phrase, or sentence which interrupts a sentence and which bears no syntactic relation to that sentence at the point of interruption’ (Trask 1991:199)—expresses more than ‘source of evidence’: it is a way of referring to one’s opinion, judgement, belief, inference, assumption, doubt, attitude and more (see Urmston 1952 and Ifantidou 1993, on their varied semantic effects).

Parentheticals in English are an open class: Dixon (2005:233–238) demonstrated that verbs of ATTENTION type and THINKING semantic types can all be used as parentheticals (provided they can take a that-complement clause and have a positive meaning). So can many other verbs, and predicative adjectives.

The meanings of parentheticals—just like with lexical verbs—are broader than those of grammatical evidentials. This is what one expects of an open class. Their use to express meanings to do with assumption and inference has been documented for a number of languages outside Europe (see Mihas 2014 for Ashéninca Perené, an Arawak language from Peru; Zhang 2014, for Ersu, a Tibeto-Burman language from China; and Forker 2014, for Hinuq, a Northeast Caucasian language).

4.3 Modal Verbs and Particles

So-called ‘modal verbs’ frequently combine reference to information source with whatever other meaning they have. In agreement with Dixon’s (2005) classification, modal verbs express secondary concepts, ‘those providing semantic modification of some other verb with which they are in a syntactic or morphological construction’ (p. 96). In many languages they are a closed subclass. Secondary verbs of the same semantic group as seem (pp. 203–205), and verbs of obligation and permission often extend to cover probability, inference, and assumption.

In Dixon’s (2005:204) words,

seem is used when the arbiter is not fully certain whether the adjectival description is appropriate, or whether the statement of the complement clause in a construction like It seems that Mary found the body or Mary seems to have found the body is correct—perhaps when there is not quite enough evidence. Appear has the same syntactic possibilities and a very similar meaning, but may imply ‘can be observed by me’ in contrast to seem ‘can be inferred by me’.

A link with information source is obvious—yet information source is an overtone of seem, rather than its only meaning.

Modal verbs may share syntactic features—such as raising—and thus form a syntactically defined subclass of verbs. The choice of modal verbs in each language is limited—they are closed subclasses. This makes it difficult to treat them on a par with lexical expression of information source through verbs, and adverbs, or parentheticals. Modal verbs tend to have more restricted meanings, close to those of grammatical evidentials. As exponents of information source, they are akin to evidentiality strategies (see § 3 above; and discussion in Aikhenvald 2004:147–148).

A plethora of particles referring to verbal report, or inference, or both may form a largish but closed class. For instance, Lithuanian has over twenty-five particles referring to verbal report or inference (Wiemer 2007). None of them is obligatory. They vary in their origin. The form girdì, literally ‘you hear’, used to mark reported information, comes from a depleted and reanalyzed verb of perception. The reported speech particle tariamai, a present passive participle of the verb ‘say, pronounce’, comes from a grammaticalized verb of speech. The meanings of these particles tend to be much less fine-grained and less specific than those of members of verbs and of nouns as open classes. This is another non-obligatory, and yet non-lexical, way of expressing information source. They can be considered a type of evidentiality strategy (see Figure 1.1).

4.4 Speech Report Constructions

Every language has a way of reporting what someone else has said. This can be cast as a direct, or an indirect speech report (see Aikhenvald 2011 for a summary). Multiclausal speech report constructions can be viewed as lexical ‘paraphrases’ of meanings grammaticalized in closed evidential systems. And in many languages, speech reports acquire epistemic overtones. They are often used to transmit something one does not really believe (see, for instance, Dimmendaal 2001, on reported speech as a ‘hedging’ device).

And it comes as no surprise that a speech report construction is a universal source for developing reported evidentials. One such grammaticalization path involves reanalysis of a biclausal quotation or reportative construction whereby the matrix clause with the verb ‘say’ and a complement clause of this verb become a single clause via the loss or reinterpretation of the subordinator (Aikhenvald 2004:273–274; 281–283). This is what we see in marker of reported speech, dizque, in Mexican and in Colombian Spanish.

Grammaticalization is a gradual process. In Italian, grammaticalization of the speech verb has just started. As Giacalone Ramat & Topadze (2007) put it, ‘the third singular form dice is frequently used in spoken Italian mostly as a marker of direct speech, but also of indirect speech, and is morphologically invariable and positionally mobile’. Parallels to this are found in Sardinian and Rumanian.

In Giacalone Ramat & Topadze’s words, ‘the parameters which allow us to describe a shift in the direction of a more grammaticalized category are: 1) decategorization (i.e. loss of inflectional distinctions), 2) positional freedom, 3) variability in scope (i.e. single constituent vs. entire clause scope), 4) semantic erosion’. And this is what we find in expressions involving diz que (literally ‘says-that’) in Latin American varieties of Spanish (Travis 2006, for Colombian Spanish; Olbertz 2007, for Mexican Spanish; Olbertz 2005 for Ecuadorian Spanish; evidence for the same phenomenon in Venezuela, Chile and Argentina come from Kany 1944), and in Brazilian Portuguese (Aikhenvald 2002, 2004).

Dizque (often written as one word) has a variable scope—as Olbertz (2007) shows, the scope of dizque can be a main clause, a subordinate clause, or a noun phrase in any syntactic function. A similar phenomenon in Colombian Spanish was described by Travis (2006). In both varieties dizque extends to cover doubt and (negative) attitude to the information and its validity—that is, it goes beyond simply reporting what someone else has said. The meanings of dizque overlap with some of the meanings of grammatical evidentials—they may involve reported speech, quotation, inference and assumption. Dizque is on the way to grammaticalizing into a marker of a category (in the sense of Heine and Kuteva 2002)—but this mechanism is very different from employing lexical items to refer to information source.

Dizque in Mexican Spanish has a greater syntactic freedom than the verb decir ‘say, speak’ it comes from, because dizque has been reanalyzed as a grammatical particle marking both speech reports and unreliable information. Dizque follows a common grammaticalization path, and is not exceptional in any way. A similar path has been documented for many other languages of the world, including Georgian. The two particles metki and -tko mark reported speech; both result from the grammaticalization of the verb tkma ‘say’ (Giacolone Ramat & Topadze 2007).

Particles marking reported speech form a closed grammatical class. They cannot be subsumed under a broad umbrella of lexical expression of information source. Instead, they can be considered evidentials in the making—akin to evidential strategies.

4.5 Expressing Information Source: A Summary

Meanings associated with information source can be expressed with members of open classes. The range of meanings is wider and more fine-grained than that of grammatical evidentials. Closed classes of particles and modal verbs tend to share their meanings with evidential strategies.

The choice of a grammatical evidential often depends on mood or tense of the clause (see § 2 above). The choice of a parenthetical or an adverb depends on what the speaker wants to say. A parenthetical, an adverb, or a modal verb can have an NP or a whole clause in its scope. For grammatical evidentials, these options are restricted. None of the means listed in § 4.1–2 forms a paradigm of any sort. In contrast, grammatical evidentials do.7

What may justify putting various verbs, adverbs and parentheticals discussed in § 4.1–2 together with modal verbs and particles is the fact that they all vaguely relate to the ways in which one knows things. All these devices for marking information source combine reference to inference, assumption, and often speech reports with increasing ‘subjectification’—a ‘historical pragmatic-semantic process whereby meanings become increasingly based in the speaker’s subjective belief state, or attitude toward what is said’ (Traugott 1996:185). This is what sets them apart from closed evidential systems—whose primary meaning has nothing to do with subjectification—and makes them similar to prototypical modalities.

5 On Terminological Clarity

Categories of grammar need to be distinguished from ‘real world’ notions. The expression of information source is akin to other conceptual notions in that they can be viewed differently, ‘in terms of their importance for the structure of the language’ (Comrie 1985:8). The first type is the set of grammatical categories. It is common knowledge that many languages have a closed set of grammaticalized expressions of location in time: these can involve present, past and future; or nonpast, recent past and remote past, etc. Along similar lines, languages with grammatical number typically distinguish singular and plural, or singular, dual and plural. These closed grammatical systems coexist with sets of lexical items which refer to location in time (e.g. yesterday, today, and so on) or to quantification (e.g. numerals and quantifiers). In addition, languages may have a potentially unlimited number of ‘composite lexical expressions’ for measuring time intervals, or for expressing quantification.

In other words, a closed grammatical system offers restricted options. This is in contrast to the lexicon where the choices are potentially open. So, for grammatical tense ‘even the maximal system would have at most tens of categories, rather than the several orders of magnitude more possible in the lexicon’ (Comrie 1985:9). The analogy with number is even more instructive: in Comrie’s (1985:9) words,

English has grammatically only a two-way opposition (singular and plural); lexically there are around thirty items (excluding those restricted to mathematical or scientific contexts); while for many speakers the possibilities for lexically composite [number] expressions are infinite.

The terminological distinction between time as a notional concept and tense as grammaticalized location in time is very handy: it helps to keep apart a closed grammatical system of tenses and a potentially open pool of temporal expressions of other sorts.8 Such terminological clarity helps understand the nature of tense and time, and sheds light on the differences between closed grammatical systems and potentially open categories of the lexicon.

Let’s follow the analogy of tense and time. In the same way as ‘tense’ refers to closed grammatical systems, ‘grammatical evidentiality’ refers to a closed set of obligatory choices of marking information source. In the same way as ‘time’ covers the potentially unlimited set of choices, ‘information source’ covers the rest. Is it appropriate to use one cover term, ‘lexical evidentiality’, to include the rest? The brief answer is ‘no’.

This approach fits in with the accepted definition of grammaticalization as anchored in opposing ‘lexical’ and ‘grammatical’ notions as basic building blocks. From a diachronic perspective, grammaticalization is conceived as ‘that part of the study of language change that is concerned with such questions as how lexical items and constructions come in certain linguistic contexts to serve grammatical functions or how grammatical items develop new grammatical functions’ (Hopper & Traugott 2003:1; Brinton & Traugott 2005:23–25; also see Heine & Kuteva 2002:4–5). In order to understand the diachronic and synchronic dynamics and development of the expression of information source, we need to adhere to terminological and conceptual clarity—rather than using the umbrella term ‘functional’ as a cover-up.

Saying that English parentheticals, or adverbs like surely or allegedly, are ‘evidentials’ is like saying that time words like yesterday or today mark tense. Any grammarian would dismiss this. But if we say that yesterday, today, two minutes ago and further similar collocations express time, and ‘present’ and ‘past’ express tense, everyone will agree.9 Along similar lines, the term ‘evidential’ is best used for closed grammatical systems, and the term ‘information source’ for the vast body of other ways of referring to ‘knowing things’.

The expression of information source which does not form a closed grammatical system and is not an extension of an existing category has been informally nicknamed ‘lexical evidentiality’. As we saw in § 4, extra-grammatical means of expressing various overtones of information source and attitude to information cover a vast ground. Some of them are purely lexical, and some belong to closed subclasses. The term ‘lexical evidentiality’ is misleading since it obscures the differences between the two types of extra-grammatical expression of information source outlined in § 4: open choices on the one hand (§ 4.1–2) and closed classes (§ 4.3–4), on the other.

The term ‘lexical evidentiality’ is confusing in yet another way. In languages with obligatory evidentials, lexical subclasses of verbs can require certain evidentiality choices. For instance, internal states and processes, felt rather than seen, are often cast in sensory (non-visual) or non-firsthand evidential. Such preferences may get lexicalized as restricted evidentiality choices for predicate types and construction types (this is a typical feature of Tibeto-Burman languages: see for instance Lidz 2007). Lexicalization of evidential choices is all too easy to confuse with ‘lexical evidentiality’.

6 Envoi

Evidentials as closed grammatical systems are different from information source marked in other ways (just like time, a real life concept, is different from tense, realized in grammar). Meanings related to information source may be expressed through open classes of verbs (of perception, opinion, speech and others), adverbs and parentheticals. These tend to be richer in their semantic range than closed systems of grammatical evidentials. Alternatively, information source may be expressed via a closed subclass of modal verbs, or via particles (often grammaticalized from verbs). These are much closer to grammatical evidentials in their nature, and their meanings.

Grammatical evidentiality is highly diffusible in language contact (see Aikhenvald 2006b, and Giacalone Ramat & Topadze 2007). And so are various other means which may involve information source. The languages of Europe share a remarkable range of semantic extensions of various verbs, parentheticals and speech reports towards expressing how one knows things.

In Bolinger’s (1991:26–27) words,

one of the happier results of recent turns in linguistics is the search for universals and the emphasis on parallel developments in various languages. […] It is as if given certain elements from a common heritage, plus a need to communicate the same ideas, common solutions are going to be hit upon sometimes, though the element of chance still plays its part. The verb parecer in Spanish and the verb seem in English reveal just such a convergence. The etymological sources are quite different. Yet once set on a path toward the common meaning of that which is evident to the senses, their developing grammars grow more and more alike.

These striking parallel developments—the essence of Sapir’s (1921:171–172) ‘parallelism in drift’—are what makes the typology of related languages, a fascinating enterprise, especially for the development of evidentials and other means of expressing knowledge and information source.











This is a substantially revised version of Aikhenvald, A.Y. 2007. Information source and evidentiality: what can we conclude?, Rivista di Linguistica 19: 1, Mario Squartini (ed.). Special issue on Evidentiality between lexicon and grammar, pp. 207–227.


This approach was adopted in Aikhenvald (2004); also see a summary in Aikhenvald (2006a). The generalizations are based on the analysis of grammars of c. 600 languages (since the publication of Aikhenvald 2004, I have had access to further grammars). I avoid limiting myself to any artificially constructed samples of languages, since these are likely to engender skewed results (also see Dixon 2010, on the pitfalls of sampling methodology).


Statements to the contrary found in Palmer (1986), van der Auwera & Plungian (1998) and Willett (1988) are not borne out by the facts of languages, and are mistaken. See the arguments in de Haan (1999), Lazard (1999, 2001) and DeLancey (2001), and a summary in Aikhenvald (2004:3–10). Some scholars whose experience is limited to a handful of familiar European languages tend to assume that evidentials are a kind of modal largely because of their absence in most major European languages, thus trying to explain an unusual category in terms of some other, more conventional, notion.


The presence of such extensions does not make evidentials into ‘modals’ (contrary to some assumptions). This can be compared to the gender systems: in many languages feminine gender is associated with diminution, or endearment (see numerous examples in Aikhenvald 2000), and masculine gender with augmentative; this however does not mean that gender is a type of diminutive or augmentative category.

Readers should be warned against gratuitously dividing languages into those where evidentials have epistemic extensions, and those where they do not (as did Plungian 2001). As shown in Chapter 5 of Aikhenvald (2004), in the same language one evidential may have an epistemic extension, and another one may not.


I have undertaken extensive fieldwork on Tariana (see, for instance, Aikhenvald 2003). I have recorded a substantial number of texts and natural conversations. All the examples from Tariana in this paper (as in my other work) come from spontaneous discourse. In my fieldwork, I avoid elicitation as being methodologically flawed (see Dixon 2010 for further fundamentals of linguistic fieldwork).


Contrary to assertions by Willett (1988) and others, an evidential may be within the scope of negation, as in Akha, a Tibeto-Burman language. An evidential can be questioned, as in Wanka Quechua. And the ‘truth value’ of an evidential may be different from that of the verb in its clause. Evidentials can be manipulated to tell a lie. One can give a correct information source and wrong information, as in saying ‘He is dead-reported’, when you were told that he is alive, or correct information and wrong information source, as in saying ‘He is alive-visual’, when in fact you were told that he is alive, and did not see him die. Two different information sources can be expressed within one clause (Aikhenvald 2004: 93; 96–98; Fleck 2007).

The grammatical category of evidentiality can be expressed through any of the affixes, clitics, or auxiliary constructions. Linguists should be warned not to take seriously generalizations based on a limited sample such as those in De Haan (2005) which provides a highly inadequate coverage of formal means of marking evidentiality.


Also see Kirsner & Thompson (1976) on a difference between ‘direct perception of a situation’ and ‘deducing a situation’ in their analysis of complements of sensory verbs in English.


In conventional linguistic terminology (Matthews 1997:263), a paradigm is defined as ‘the forms of a given noun, verb, etc. arranged systematically according to their grammatical features’ (italic mine: A.A.).


For similar distinctions in terminological traditions other than the English, see Jespersen (1924:255).


This is similar to how the linguistic literature on gender as a grammatical category does not discuss words for ‘man’ and ‘woman’, or ‘bull’ and ‘cow’ in each particular language. ‘Evidentiality in English’ has the same status as ‘gender in Hungarian’, or in Estonian. Of course, sex distinctions can be expressed in Hungarian and in Estonian if one wants to, but there is no grammatical category of gender. One can indicate information source in English, if necessary. But this is not grammatical evidentiality.


  • Aikhenvald, Alexandra Y. 2000. Classifiers: a typology of noun categorization devices. Oxford: Oxford University Press.

  • Aikhenvald, Alexandra Y. 2002. Language Contact in Amazonia. Oxford: Oxford University Press.

  • Aikhenvald, Alexandra Y. 2003. A Grammar of Tariana, from Northwest Amazonia. Cambridge: Cambridge University Press.

  • Aikhenvald, Alexandra Y. 2004. Evidentiality. Oxford: Oxford University Press.

  • Aikhenvald, Alexandra Y. 2006a. ‘Evidentiality in grammar’, in Keith Brown, ed. (2006). Encyclopedia of Language and Linguistics. 2nd edition. Volume 4. Oxford: Elsevier. 320–325 (article 0252).

  • Aikhenvald, Alexandra Y. 2006b. ‘Grammars in contact: a cross-linguistic perspective’, in Alexandra Y. Aikhenvald & R.M.W. Dixon, eds. (2006). Grammars in Contact: A Cross-Linguistic Typology. Oxford: Oxford University Press, 1–66.

  • Aikhenvald, Alexandra Y. 2010. Imperatives and commands. Oxford: Oxford University Press.

  • Aikhenvald, Alexandra Y. 2011. ‘Speech reports: a cross-linguistic perspective’, in Language at large. Essays on syntax and semantics, by Alexandra Y. Aikhenvald and R.M.W. Dixon. Leiden: Brill, 290–326.

  • Aikhenvald, Alexandra Y. 2012. Languages of the Amazon. Oxford: Oxford University Press.

  • Aikhenvald, Alexandra Y. 2014. ‘The grammar of knowledge: a cross-linguistic view of evidentials and the expression of information source’, in Aikhenvald and Dixon (eds.), 1–51.

  • Aikhenvald, Alexandra Y. 2018. ‘Evidentiality: The framework’, in Alexandra Y. Aikhenvald (ed.) (2018). The Oxford Handbook of Evidentiality. Oxford: Oxford University Press, 1–46.

  • Aikhenvald, Alexandra Y. & R.M.W. Dixon. eds. 2003. Studies in Evidentiality. Amsterdam: John Benjamins.

  • Aikhenvald, Alexandra Y. & R.M.W. Dixon. eds. 2014. The grammar of knowledge: a cross-linguistic typology. Oxford: Oxford University Press.

  • Babel, Anna. 2009. ‘Dizque, evidentiality and stance in Valley Spanish’. Language in Society 38: 487–511.

  • Boas, Franz. 1911. ‘Kwakiutl’, in Franz Boas, ed. (1911), Handbook of American Indian Languages, Part 1. Smithsonian Institution, Bureau of American Ethnology Bulletin 40, 423–557.

  • Boas, Franz. 1938. ‘Language’, in Franz Boas, ed. (1938). General Anthropology. Boston, New York: D.C. Heath and Company. 124–145.

  • Bolinger, Dwight. 1991. Essays on Spanish: Words and Grammar. Newark, Delaware: Juan de la Cuesta.

  • Brinton Laurel J. & Elizabeth C. Traugott (2005). Lexicalization and Language Change. Cambridge: Cambridge University Press.

  • Chafe, Wallace L. & Johanna Nichols, eds. 1986. Evidentiality: The Linguistic Coding of Epistemology, Norwood, NJ: Ablex.

  • Chirikba, Viacheslav. 2003. ‘Evidential category and evidential strategy in Abkhaz’, in Aikhenvald & Dixon (2003:243–272).

  • Comrie, Bernard 1976. Aspect. Cambridge: Cambridge University Press.

  • Comrie, Bernard 1985. Tense. Cambridge: Cambridge University Press.

  • De Haan, Ferdinand. 1999. ‘Evidentiality and epistemic modality: setting boundaries’. Southwest Journal of Linguistics 18: 83–102.

  • De Haan, Ferdinand. 2005. ‘Semantic distinctions of evidentiality’, in Haspelmath Martin et al. (eds. 2005). World Atlas of Language Structures. Berlin: Mouton de Gruyter, 318–321.

  • DeLancey, Scott. 2001. ‘The mirative and evidentiality’. Journal of Pragmatics 33: 369–382.

  • Dendale, P. 1993. ‘Le conditionnel de l’ information incertaine: marqueur modal ou marqueur évidentiel?’ pp. 165–176 XXe Congrès International de Linguistique et Philologie Romanes, Tome I, Section I. La phrase, edited by Gerold Hilty. Tübingen: Francke.

  • Dendale, P. and Julie Van Bogaert. 2007. ‘A semantic description of French lexical evidential markers and the classification of evidentials’. Rivista Italiana di Linguistica 19: 65–90.

  • Dimmendaal Gerrit. 2001. ‘Logophoric marking and represented speech in African languages as evidential hedging strategies’. Australian Journal of Linguistics 21: 131–157.

  • Dixon R.M.W. 2003. ‘Evidentiality in Jarawara’, in Aikhenvald & Dixon (2003:165–187).

  • Dixon R.M.W. 2005. A Semantic Approach to English Grammar. Oxford: Oxford University Press.

  • Dixon R.M.W. 2010. Basic linguistic theory. Volume 1. Methodology. Oxford: Oxford University Press.

  • Eberhard, David. 2009. A grammar of Mamaindê (Nambiquara). Amsterdam: LOT.

  • Eberhard, David. 2018. ‘Evidentiality in Nambikwara languages’, in Alexandra Y. Aikhenvald (ed.) (2018). The Oxford Handbook of Evidentiality. Oxford: Oxford University Press, 333–356.

  • Fleck, David. 2007. ‘Evidentiality and double tense in Matses’. Language 83: 589–614.

  • Forker, Diana. 2014. ‘The grammar of knowledge in Hinuq’, in Aikhenvald and Dixon (eds.), 52–68.

  • Giacalone Ramat, Anne & Manana Topadze. 2007. ‘The coding of evidentiality: a comparative look at Georgian and Italian’. Rivista Italiana di Linguistica 19: 109–128.

  • Gronemeyer, C. 1997. ‘Evidentiality in Lithuanian’. Working Papers 46: 93–112. Lund University, Department of Linguistics.

  • Heine, Bernd & Tania Kuteva. 2002. World Lexicon of Grammaticalization. Cambridge: Cambridge University Press.

  • Hewitt B. George. 1995. Georgian Grammar. Amsterdam: John Benjamins.

  • Hopper, Paul J. & Elizabeth C. Traugott. 2003. Grammaticalization. Cambridge: Cambridge University Press.

  • Ifantidou, Elly. 1993. ‘Parentheticals and relevance’. UCL Working Papers in Linguistics 5: 193–210.

  • Jacobsen William H. Jr. 1986. ‘The heterogeneity of evidentials in Makah’, in Chafe & Nichols (1986:3–28).

  • Jakobson, Roman O. 1957. ‘Shifters, verbal categories, and the Russian verb’. Cambridge: Harvard University.

  • Jespersen, Otto. 1924. The Philosophy of Grammar. London: George Allen & Unwin Ltd.

  • Kany Charles. 1944. ‘Impersonal dizque and its variants in American Spanish’. Hispanic Review 12: 168–177.

  • Kirsner, R.S. & Sandra A. Thompson. 1976. ‘The role of pragmatic inference in semantics: a study of sensory verb complements in English’. Glossa 10: 200–240.

  • Lazard, Gilbert. 1957. ‘Caractères distinctifs de la langue tadjik’. Bulletin de la Société de Linguistique de Paris 52: 117–186.

  • Lazard, Gilbert. 1999. ‘Mirativity, evidentiality, mediativity, or other?’ Linguistic Typology 3: 91–110.

  • Lazard, Gilbert. 2001. ‘On the grammaticalization of evidentiality’. Journal of Pragmatics 33: 359–367.

  • Lidz, Liberty A. 2007. ‘Evidentiality in Yongning Na (Mosuo).’ Linguistics of the Tibeto-Burman Area 30:45–88.

  • Lowe, I. 1999. ‘Nambiquara’, pp. 269–292 of The Amazonian languages, edited by R.M.W. Dixon and Alexandra Y. Aikhenvald. Cambridge: Cambridge University Press.

  • Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.

  • Mathiassen, T. 1996. A Short Grammar of Lithuanian. Columbus, OH: Slavica Publishers.

  • Matthews, P.H. 1997. Oxford Dictionary of Linguistics. Oxford: Oxford University Press.

  • Mihas, Elena. 2014. ‘Expression of information source meanings in Ashéninca Perené’, in Aikhenvald and Dixon (eds.), 209–226.

  • Olbertz, Hella. 2005. ‘Dizque en el español andino ecuatoriano: conservador e innovador’, in Hella Olbertz & Pieter Muysken, eds. (2005), Encuentros y conflictos. Bilingüismo y contacto de lenguas en el mundo andino. Madrid: Iberoamericana, 77–94.

  • Olbertz, Hella. 2007. ‘Dizque in Mexican Spanish: the subjectification of reportative meaning’. Rivista di Linguistica 19: 1: 151–172.

  • Palmer, F.R. 1986. Mood and Modality. Cambridge: Cambridge University Press.

  • Pietrandrea, Paola. 2007. ‘The grammatical nature of some epistemic-evidential adverbs in spoken Italian.’ Rivista Italiana di Linguistica 19: 39–63.

  • Plungian, Vladimir A. 2001. ‘The place of evidentiality within the universal grammatical space’. Journal of Pragmatics 33: 349–358.

  • Pusch, Claus D. 2007. ‘Is there evidence for evidentiality in Gascony Occitan?’ Rivista Italiana di Linguistica 19: 91–108.

  • Robins, R. 1967. A Short History of Linguistics. London: Longmans.

  • Sapir, Edward. 1921. Language. New York: Harcourt, Brace and World.

  • Squartini, Mario. 2007. ‘Lexical vs grammatical evidentiality in French and Italian’. Linguistics 46: 917–947.

  • Storch, Anne. 2018. ‘Evidentiality and the expression of knowledge: an African perspectve’, in Alexandra Y. Aikhenvald (ed.) (2018). The Oxford Handbook of Evidentiality. Oxford: Oxford University Press, 610–628.

  • Storch, Anne and Jules Jacques Coly. 2014. ‘The grammar of knowledge in Maaka (Western Chadic, Nigeria)’, in Aikhenvald and Dixon (eds.), 190–208.

  • Timberlake, Alan. 1982. ‘The impersonal passive in Lithuanian’. Proceedings of the Eighth Annual Meeting of the Berkeley Linguistics Society, 508–523.

  • Trask, R.L. 1991. A Dictionary of Grammatical Terms in Linguistics. London: Routledge.

  • Traugott, Elizabeth Closs. 1996. ‘Subjectification and the development of epistemic meaning: the case of promise and threaten’, in Toril Swan & Olaf J. Westvik, eds. (1996). Modality in Germanic Languages: Historical and Comparative Perspectives. Berlin: Mouton de Gruyter, 185–210.

  • Travis, Catherine. 2006. ‘Dizque: a Colombian evidentiality strategy’. Linguistics 44: 1269–1297.

  • Urmston, J.O. 1952. ‘Parenthetical verbs’. Mind 61: 480–496.

  • Valenzuela, Pilar. 2003. ‘Evidentiality in Shipibo-Konibo, with a comparative overview of the category in Panoan’, in Aikhenvald & Dixon (2003:33–62).

  • van der Auwera, Johan & Vladimir A., Plungian. 1998. ‘On modality’s semantic map’. Linguistic Typology 2: 79–124.

  • Weber, David J. 1986. ‘Information perspective, profile, and patterns in Quechua’, in Chafe & Nichols (1986:137–155).

  • Wiemer, Björn. 2007. ‘Lexical markers of evidentiality in Lithuanian’. Rivista Italiana di Linguistica 19: 173–208.

  • Willett, Thomas. 1988. ‘A cross-linguistic survey of the grammaticization of evidentiality’. Studies in Language 12: 51–97.

  • Zhang, Sihong. 2014. ‘The expression of knowledge in Ersu’, in Aikhenvald and Dixon (eds.), 89–107.


All Time Past Year Past 30 Days
Abstract Views 251 95 5
Full Text Views 14 5 0
PDF Views & Downloads 22 11 0