Evidentiality

In linguistics, evidentiality is, broadly, the indication of the nature of evidence for a given statement; that is, whether evidence exists for the statement and if so, what kind. An evidential is the particular grammatical element that indicates evidentiality. Languages with only a single evidential have had terms such as mediative, médiatif, médiaphorique, and indirective used instead of evidential.
Evidentiality may be direct or indirect: direct evidentials are used to describe information directly perceived by the speaker through vision as well as other sensory experiences while indirect evidentials consist of the other grammatical markers for evidence such as quotatives and inferentials.

Introduction

All languages have some means of specifying the source of information. European languages often use modal verbs or other lexical words or phrases.
Some languages have a distinct grammatical category of evidentiality that is required to be expressed at all times. In contrast, the elements in European languages indicating the information source are optional and usually do not indicate evidentiality as their primary function; thus, they do not form a grammatical category. The obligatory elements of grammatical evidentiality systems may be translated into English, variously, as I hear that, I see that, I think that, as I hear, as I can see, as far as I understand, they say, it is said, it seems, it seems to me that, it looks like, it appears that, it turns out that, alleged, stated, allegedly, reportedly, obviously, etc.
Alexandra Aikhenvald reports that about a quarter of the world's languages have some type of grammatical evidentiality. Laura Mazzoni has since conducted a preliminary study on evidentiality in Italian Sign Language.
Grammatical evidentiality may be expressed in different forms depending on the language, such as through affixes, clitics, or particles. For example, Japanese has inferential evidentials and reportive markers that are realized as suffixes on a variety of mainly verbal predicates, and as grammaticalized nouns. As another example, Eastern Pomo uses four evidential suffixes that are added to verbs: -ink’e, -ine, -·le, and -ya.

Evidential type	Example verb	Gloss
nonvisual sensory	pʰa·békʰ-ink’e	"burned"
inferential	pʰa·bék-ine	"must have burned"
hearsay	pʰa·békʰ-·le	"burned, they say"
direct knowledge	pʰa·bék-a	"burned"

Many languages with grammatical evidentiality mark evidentiality independently from tense-aspect or epistemic modality, which is the speaker's evaluation of the information, i.e. whether it is reliable, uncertain, probable.
The use of evidentiality has pragmatic implications. In languages that do not mark evidentiality distinctly from epistemic modality, for example, a person who makes a false statement qualified as a belief may be considered mistaken, while a person who makes a false statement qualified as a personally observed fact will probably be considered to have lied. More generally, a speaker of a language that does have obligatory grammatical evidentiality is required to cognitively engage with the source of their belief of any statement in a manner that the speaker of languages without obligatory evidentiality may gloss over.
In some languages, evidential markers also serve other purposes, such as indicating the speaker's attitude towards, or belief in, the statement. Usually a direct evidential marker may serve to indicate that the speaker is certain about the event stated. Using an indirect evidential marker, such as one for hearsay or reported information, may indicate that the speaker is uncertain about the statement, or doesn't want to take responsibility for its truth. A "hearsay" evidential may then have the undertone of "that's what they say; whether or not it's true is nothing I can take responsibility for". In other languages, this is not the case. Therefore, one should distinguish between such evidential markers that only mark source of knowledge, and such evidential markers that serve other functions, such as marking epistemic modality.
Evidentials can also be used to "deflect culpability" in a statement. In his dissertation on Nanti, a Peruvian Amazonian language, Lev Michael refers to an example in which a young girl is accidentally burned, and a community member questions her mother about how it happened. Her mother uses the evidential marker ka which translates to "presumably," to deflect responsibility for the girl's mistake.
Some languages are borderline cases. For example, the Romance languages are mostly like English in not having grammatical evidentiality, but do have a conditional mood which has three uses: conditions, future-in-the-past, and hearsay. Thus in journalistic French, there is frequently a distinction between Il a reconnu sa culpabilité and Il aurait reconnu sa culpabilité: both translate to "He has admitted his guilt," but with an implication of certainty with the first, and the idea of "reportedly" with the second. The same happens in Spanish and in Portuguese.
Aikhenvald identified five semantic categories that recurrently occur across languages of the world:

Visual Sensory
Non-Visual Sensory
Inferentials
Hearsay Reportatives
Quotative Reportatives

No language has been reported to have special forms for smell, taste or feeling although these may be covered by non-visual evidentials.

Types according to Aikhenvald

Following the typology of Alexandra Aikhenvald, there are two broad types of evidential marking:

indirectivity marking
evidential marking

The first type indicates whether evidence exists for a given statement, but does not specify what kind of evidence. The second type specifies the kind of evidence.

Indirectivity (type I)

Indirectivity systems are common in Uralic and Turkic languages. These languages indicate whether evidence exists for a given source of information; thus, they contrast direct information and indirect information. Unlike the other evidential "type II" systems, an indirectivity marking does not indicate information about the source of knowledge: it is irrelevant whether the information results from hearsay, inference, or perception; however, some Turkic languages distinguish between reported indirect and non-reported indirect, see Johanson 2003, 2000 for further elaboration. This can be seen in the following Turkish verbs:
In the word geldi, the unmarked suffix -di indicates past tense. In the second word gelmiş, the suffix -miş also indicates past tense but indirectly. It may be translated into English with the added phrases 'obviously', 'apparently' or 'as far as I understand'. The direct past tense marker -di is unmarked in the sense that whether or not evidence exists supporting the statement is not specified.

Evidentiality (type II)

The other broad type of evidentiality systems specifies the nature of the evidence supporting a statement. These kinds of evidence can be divided into such categories as:
Sensory evidentials can often be divided into different types. Some languages mark visual evidence differently from nonvisual evidence that is heard, smelled, or felt. The Kashaya language has a separate auditory evidential.
An inferential evidential indicates information was not personally experienced but was inferred from indirect evidence. Some languages have different types of inferential evidentials. Some of the inferentials found indicate:
In many cases, different inferential evidentials also indicate epistemic modality, such as uncertainty or probability. For example, one evidential may indicate that the information is inferred but of uncertain validity, while another indicates that the information is inferred but unlikely to be true.
Reportative evidentials indicate that the information was reported to the speaker by another person. A few languages distinguish between hearsay evidentials and quotative evidentials. Hearsay indicates reported information that may or may not be accurate. A quotative indicates the information is accurate and not open to interpretation, i.e., is a direct quotation. An example of a reportative from Shipibo :

Typology of evidentiality systems

The following is a brief survey of evidential systems found in the languages of the world as identified in Aikhenvald. Some languages only have two evidential markers while others may have six or more. The system types are organized by the number of evidentials found in the language. For example, a two-term system will have two different evidential markers; a three-term system will have three different evidentials. The systems are further divided by the type of evidentiality that is indicated. Languages that exemplify each type are listed in parentheses.
The most common system found is the A3 type.
Two-term systems:

A1. witness, nonwitness
A2. nonfirsthand, everything else
A3. reported, everything else

Three-term systems:

B1. visual sensory, inferential, reportative
B2. visual sensory, nonvisual sensory, inferential
B3. nonvisual sensory, inferential, reportative
B4. witness, nonwitness, inferential, reportative

Four-term systems:

C1. visual sensory, nonvisual sensory, inferential, reportative
C2. visual sensory, inferential #1, inferential #2, reportative
C3. nonvisual sensory, inferential #1, inferential #2, reportative
C4. visual sensory, inferential, reportative #1, reportative #2
C5. witness, inferential, renarrative, dubitative

Five-plus term systems:

visual sensory, nonvisual sensory, inferential, reportative, assumed
witness, inferential, reportative, assumed, "internal support"
visual sensory, nonvisual sensory, inferential, reported, heard from known source, direct participation
nonvisual sensory, inferential #1, inferential #2, inferential #3, reportative
inferential, anticipation, performative, deduction, induction, hearsay, direct observation, opinion, assumed, "to know by culture", "to know by internal"