320 Event-Based

Rothstein S (ed.) (1998). Events and grammar. Dordrecht: Tenny C & Pustejovsky J (eds.) (2000). Events as grammati- Kluwer Academic Publishers. cal objects. Stanford, CA: CSLI Publications. Schein B (1993). and events. Cambridge, MA: MIT Vlach F (1983). ‘On for perception.’ Press. Synthese 54, 129–152.

Evidentiality in Grammar A Y Aikhenvald, La Trobe University, (1) Juse irida di-manika-ka Bundoora, Australia Jose´ football 3person.masculine. ß 2006 Elsevier Ltd. All rights reserved. singular-play- RECENT.PAST. VISUAL Evidentiality is a that has ‘Jose´ played football (we saw it)’ source of information as its primary meaning – wheth- er the narrator actually saw what is being described, If one just heard the noise of a football game but could or made inferences about it based on some evidence, not see what was happening, (2) is the thing to say: or was told about it, etc. Languages vary in how many (2) Juse irida di-manika-mahka information sources have to be marked. Many just Jose´ football 3person.masculine. mark information reported by someone else; others singular-play- distinguish firsthand and nonfirsthand information RECENT.PAST.NONVISUAL sources. In rarer instances, visually obtained data are ‘Jose´ played football (we heard it)’ contrasted with data obtained through hearing and If one sees that the football is not in its normal smelling, and through various kinds of inference. place in the house, Jose´ and his football boots are As Boas (1938: 133) put it, ‘‘while for us definite- gone, with crowds of people coming from the football ness, number, and time are obligatory aspects, we find ground, these details are enough for us to infer that in another language location near the speaker or Jose´ is playing football: somewhere else, source of information – whether seen, heard, or inferred – as obligatory aspects.’’ The (3) Juse irida di-manika-nihka terms ‘verificational’ and ‘validational’ are sometimes Jose´ football 3person.masculine. singular-play- used in place of ‘evidential.’ French linguists employ RECENT.PAST.INFERRED the term ‘mediative’ (Guentche´va, 1996). The term ‘Jose´ played football (we infer it from visual ‘evidential’ was first introduced by Jakobson (1957). evidence)’ A summary of work on recognizing this category, and naming it, is in Jacobsen (1986) and Aikhenvald If Jose´ is not at home on a Sunday afternoon, we (2004). can safely say (4). Our inference is based on gene- Evidentiality is a verbal grammatical category in its ral knowledge about Jose´’s habits: he usually plays own right, and it does not bear any straightforward football on Sunday afternoon. relationship to truth, the validity of a statement, or (4) Juse irida di-manika-sika the speaker’s responsibility. Neither is evidentiality a Jose´ football 3person.masculine. subcategory of epistemic or any other modality (pace singular-play- Palmer 1986: 51): in numerous languages irrealis and RECENT.PAST.ASSUMED other modalities can occur together with evidentials ‘Jose´ played football (we infer it from general (also see discussion in De Haan, 1999; Lazard, 1999, knowledge)’ 2001; and DeLancey, 2001). If one learnt the information from someone else, In Tariana, an Arawak language spoken in the then (5) – with a reported evidential – is the only multilingual area of the Vaupe´s in northwest Amazo- correct option: nia, speakers have to specify whether they saw the event happen, or heard it, or know about it because (5) Juse irida di-manika-pidaka somebody else told them, etc. Omitting an eviden- Jose´ football 3person.masculine.singular- tial typically results in an ungrammatical and highly play-REC.P.REP unnatural sentence (see details in Aikhenvald, 2004). ‘Jose´ played football (we were told)’ If one saw Jose´ play football, (1) would be Languages that have ‘evidentiality’ as a grammati- appropriate: cal category vary in how many types of evidence they Evidentiality in Grammar 321 mark. Some distinguish just two terms (eyewitness A2. Nonfirsthand and everything else. The nonfirst- and noneyewitness, or reported and everything else), hand evidential covers a large domain of infor- while others six or more terms. mation acquired through senses other than Every language has some lexical way of referring seeing, and through hearsay and inference of all to information source, e.g., English reportedly or sorts, as in many Caucasian languages, and also allegedly. Such lexical expressions may become gram- in Turkic and Finno-Ugric languages (Johanson, maticalized as evidential markers. Nonevidential cate- 2003; Aikhenvald, 2003a). Forms unmarked for gories may acquire a secondary meaning relating to evidentiality are evidentially neutral (they do not information source. Conditionals and other nonde- have any to information source). The clarative moods may acquire overtones of uncertain nonfirsthand evidential in Abkhaz, a Northwest information obtained from some other source, for Caucasian language (Chirikba, 2003), can describe which the speaker does not take any responsibility; inference, as in (7): that the woman was crying is the best known example is the French conditional inferred from the fact that her eyes are red. The (Dendale, 1993). and perfects acquire over- same form is used for reported information. tones of nonfirsthand information in many Iranian " "o and , and resultative nominalizations (7) je-q a-n d ewa-zaaren it-be-PAST (s)heþcry-NONFIRSTHAND and passives (also with a resultative meaning) can ‘(when she came up to the light, to the fire, express similar meanings. In other languages, the her eyes were very red) Apparently, she choice of a complementizer or a type of had been crying’ (speaker’s inference) may serve to express meanings related to how one knows a particular fact. In English, different A3. Reported (or ‘hearsay’) and everything else. Sys- complement distinguish an auditory and a tems of this sort with one, reported, evidential, hearsay meaning of the verb hear:sayingIheardBrazil which covers information acquired through beating France implies actual hearing, while I heard someone else’s narration, are widespread all that Brazil beat France implies a verbal report of the over the world (see, for instance, Silver and result. These evidential-like extensions are known as Miller (1997: 38) on North American Indian ‘evidentiality strategies’ (Aikhenvald, 2003a). Histori- languages). (8) comes from Estonian: cally, they may give rise to grammatical evidentials. (8) Ta olevat Languages with evidentials fall into a number of he be.REPORTED subtypes, depending on how many information arsti-teaduskonna sources acquire distinct grammatical marking. Small doctor-faculty.GENITIVE.SINGULAR systems with just two choices cover: lo˜ peta-nud A1. Firsthand vs. nonfirsthand. The firsthand term finish-PAST.PARTICIPLE ‘He is said to have completed his studies of typically refers to information acquired through medicine (but I wouldn’t vouch for it)’ vision (or hearing, or other senses), and the non- firsthand covers all other sources, including in- A4. Sensory evidence and reported. The sensory evi- formation acquired by senses other than seeing, dential refers to something one has seen, heard, by inference and by verbal report. A useful over- smelt, or felt (by touch); the other evidential refers view of such systems, especially in Turkic and to verbal report, as in the Australian languages Iranian languages, is in Johanson and Utas Ngiyambaa (Wangaaybuwan-Ngiyambaa) and (2000). In (6), from Jarawara, an Arawa´ lan- Diyari (Dieri). guage from Brazil (Dixon, 2003), a firsthand Of these, A1 and A4 are clear-cut two-term systems, evidential marks what the speaker could see, while A2 and A3 include an ‘everything else,’ or and the nonfirsthand refers to what he could evidentially neutral, term. not see Systems with three choices are: (6) Wero kisa-me-no, B1. Direct (or visual), Inferred, and Reported. name get.down-BACK-IMMEDIATE.PAST. Depending on the system, the first term can NONFIRSTHAND.masculine ka-me-hiri-ka refer to visually acquired information, as in be.in.motion-BACK- Qiang, a Tibeto-Burman language from China; RECENT.PAST.FIRSTHAND.masculine- or to information based on sensory evidence, DECLARATIVE.masculine which covers seeing, hearing, smelling and ‘Wero got down from his hammock (which touching something, as in Mosete´n, an isolate I didn’t see), and went out (which I did see)’ from Bolivia, the Jaqi, or Aymara languages 322 Evidentiality in Grammar

from Bolivia (Hardman, 1986), and Shilluk (a form, its evidentiality value is typically recoverable Western Nilotic language from Sudan). Quechua from the . languages have three evidentiality specifications: Four-term systems cover: direct evidence (-mi), conjectural (-chi,-chr(a)) C1. Visual, Nonvisual sensory, Inferred, Reported, as and reported (-shi) (see Floyd, 1999, on Wanka in numerous East spoken in Quechua: northwest Amazonia, and in Eastern Pomo, a (9) trabaja-an˜a-m li-ku-n Pomoan language from California. work-PURPOSE. go-REFLEXIVE-3person C2. Visual or direct evidence, Inferred, Assumed, MOTION-now- REPORTED, as in Shipibo-Konibo, a Panoan lan- DIRECT. guage from Peru. The sensory evidential in a C2 EVIDENTIAL system can refer to firsthand knowledge ‘He’s gone to work’ (I saw him go) acquired through any physical sense, be it vision, (10) chay lika-a-nii hearing, smell, taste, or touch, as in Shipibo- that see-NOMINALISER-1person Konibo; or it may refer just to information juk-ta-chra-a lika-la acquired by seeing, as in Tsafiki, a Barbacoan other-ACCUSATIVE- see-PAST language from Ecuador (Dickinson, 2000). The CONJECTURAL. visual evidential is formally unmarked; there is EVIDENTIAL-TOPIC one suffix marking information inferred from ‘The witness (‘my seer’) must have seen direct physical evidence, another for inference someone else’ (her house was robbed; she from general knowledge, and an additional one saw someone next to her house, it was not me, I infer it was (-chr) someone else) for reported, or hearsay. (11) Ancha-p-shi (12) Manuel ano fi-e too.much-GENITIVE-REPORTED.EVIDENTIAL Manuel food eat-DECLARATIVE wa"a-chi-nki ‘Manuel ate’ (the speaker saw him) cry-CAUSATIVE-2person (13) Manuel ano fi-nu-e wamla-a-ta Manuel food eat-INFERRED- girl-1person-ACCUSATIVE DECLARATIVE ‘You make my daughter cry too much’ (they ‘Manuel ate’ (the speaker sees the dirty dishes) tell me) (14) Manuel ano fi-n-ki-e B2. Visual, Nonvisual sensory, Inferred are found Manuel food eat-NOMINALISER- in Washo, from the California-Nevada border VERB.CLASS:do- and in Siona, a West Tucanoan language from DECLARATIVE Ecuador. ‘Manuel ate’ (He always eats at 8:00 and it’s B3. Visual, Nonvisual sensory, Reported are found in now 9:00) Oksapmin (isolate from Papua New Guinea), C3. Direct (or visual) Inferred, Reported, Quotative Maricopa, a Yuman language from Arizona, are found in Cora, a Uto-Aztecan language from and Dulong, a Tibeto-Burman language from Mexico. Burma. B4. Nonvisual sensory, Inferred, Reported are found Four-term systems involve at least one sensory in the Samoyedic languages Nganasan and Enets. specification. If there is just one sensory evidential, B5. Reported, Quotative and everything else. Only additional complexity arises within inferred eviden- reported and quoted information requires a spe- tials (as in C2: one evidential then refers to inference cial marker in a few North American Indian based on visible results, and the other one to infer- languages, e.g., Comanche, a Uto-Aztecan lan- ence based on reasoning and assumption). Additional guage from Oklahoma. choices between reported evidentials involve distinguishing reported and quoted information (C3). These systems include at least one sensory spec- The only type of multiterm system found in more ification. The nonvisual sensory evidential in B2, than one language involves: B3, and B4 systems typically covers information acquired by hearing, smelling, and touching, and D1. Visual, Nonvisual sensory, Inferred, Assumed, feeling. The inferred evidential refers to inference and Reported. This system was exemplified in based on visible traces and assumption, while the (1)–(5) above and is also found in Tucanoan reported evidential describes any kind of verbal languages in Brazil and Colombia, such as report. If a language has an evidentially unmarked Tuyuca and Desano (Barnes, 1984). Evidentiality in Grammar 323

Systems with more than five terms have just two man, one uses a nonvisual evidential, while the story sensory evidentials, and a number of evidentials is told in reported evidential because the narrator based on inference and assumption of different heard it from someone else. These features make kinds, as in the Nambiquara languages from Brazil, evidentiality similar to a predication in its own and Foe and Fasu, of the Kutubuan family spoken in right. Further arguments to the same effect include: the Southern Highlands of Papua New Guinea. In . An evidential may be within the of nega- some languages, a wide variety of evidential mean- tion, as in Akha, a Tibeto-Burman language ings may be expressed in different slots of the verbal (Hansson, 2003). In (16), the visual experience word or within a clause. Different evidentiality spe- and not the verb itself is being negated: cifications are ‘scattered’ throughout the grammar, and by no means form a unitary category, as in (16) a`jOqa´N dı` Makah, a Wakashan language from Washington he NOUN.PARTICLE beat State (Jacobsen, 1986), in Eskimo languages and O a`shu´ Wa` ma` Na´ in Western Apache, an Athabaskan language from VERBAL.PARTICLE who not VISUAL ‘I do not know/can’t see who is beating him’ Arizona (de Reuse, 2003: 97). Evidentials can be expressed with a wide array of . An evidential can be questioned, as in Wanka Que- morphological mechanisms and processes. There is chua (Floyd, 1999: 132). no correlation between the existence of evidentials . The ‘truth value’ of an evidential may be different and language type. Even pidgins and creoles are from that of the verb in its clause. Evidentials can known to have had evidentials (as did Chinese Pidgin be manipulated to tell a lie. One can give a correct Russian). Examples of a truly functionally unmarked information source and wrong information, as form in an evidentiality system are rare. The first- in saying ‘He is dead-REPORTED,’ when you were hand, visual, or a combined visual and sensory, evi- told that he is alive, or correct information and dential tends to be less formally marked than any wrong information source, as in saying ‘He is other term. This term is formally unmarked in some alive-VISUAL,’ when in fact you were told that he is languages, as we saw in (12), from Tsafiki. Evidenti- alive, and did not see him die. ality neutral terms are a property of a few systems . And finally, an evidential can have its own time where an evidential is opposed to ‘everything else’ reference, distinct from the time reference of the (these are A2, A3, A5, and B5). This is quite different event talked about (see Aikhenvald, 2003b, for from omitting an evidential, which can happen either Tariana). if the information source is clear from the context, or Evidentials vary in their semantic extensions, if evidentials are mutually exclusive with some other depending on the system and its structure. The first- morpheme, e.g., mood, as in Samoyedic languages. hand term in two term-systems typically refers to Cooccurrence of different evidentials in one clause visual and often other sensory information, and can – and the different morphological statuses of eviden- be extended to denote the direct participation, con- tials – provides a tool for distinguishing evidentiality trol, and volitionality of the speaker. The sensory subsystems within one language. If a language has evidential in A4 systems refers to sensory perception several distinct evidentiality subsystems, the reported of any kind, without any epistemic or other over- specification is most likely to be set apart from others. tones. The nonfirsthand term in A1 and A2 systems Evidentials differ from other grammatical cate- means the opposite of firsthand. The nonfirsthand gories in a number of ways. The information source often implies lack of participation, lack of control, can be marked more than once in a clause. Two nonspecific evidence (or no evidence at all), inference, sources can be different, but somehow linked and hearsay. An to hearsay is sometimes together, as in Tsafiki (Dickinson, 2000: 408): found but is not universal. (15) Manuel ano fi-nu-ti-e There are hardly any epistemic extensions in A1 Manuel food eat-INFERENCE. evidentiality systems with two choices. Languages PHYSICAL.EVIDENCE- tend to have other ways of expressing probability REPORTED- and possibility. DECLARATIVE In systems with three or more terms, the visual ‘He said/they say Manuel has eaten’ (they didn’t or the direct evidential usually covers information see him, but they have direct physical acquired through seeing, and also generally known evidence) and observable facts. It may be extended to indicate In Eastern Pomo, the two sources can be fully certainty. The nonvisual sensory evidential in B2, B3, distinct: describing information source of a blind and B4 systems refers to information acquired by 324 Evidentiality in Grammar hearing, smell, touch, or feeling (such as an injection), 2. The most frequent evidential in commands is and has no epistemic extensions. No language has a reported (‘do what someone else told you to’). special evidential to cover smell but not auditory The choice of an evidential in questions may con- information. The inferred evidential typically covers tain reference to the source of information avail- inference based on visual evidence, on nonvisual sen- able to the speaker, to the addressee or to both. sory evidence, on reasoning or on assumption. It is 3. Fewer evidentials may be used in negative clauses also used to refer to someone else’s ‘internal states’– than in positive. feelings, knowledge, and the like. It may acquire an 4. Nonindicative modalities (conditional, dubitative epistemic extension of ‘conjecture,’ uncertainty, and and so on) may allow fewer evidential lack of control. specifications than the indicative. In many lan- The reported evidential is semantically uniform in guages, evidentials may not be used in future systems of all types. Its core meaning is to mark that which is, by its nature, a kind of modality. information comes from someone else’s report. 5. The maximum number of evidential specifications A reported evidential can be used as a quotative, to is expected in past tenses. In some languages, as in indicate the exact authorship of the information, or to Jarawara (Dixon, 2003), firsthand and nonfirst- introduce a direct quote. It can be used for a second- hand evidentials are distinguished only in the past. hand or thirdhand report. A reported evidential may The source of information for an event is often develop an epistemic extension of unreliable informa- based on its result, hence the link between first- tion, as a means of ‘’ responsibility for the hand/nonfirsthand, on the one hand, and past, information to some other source one does not perfect, perfective, and resultative on the other. vouch for, as in Estonian: example (8) has overtones Evidentials often come from grammaticalized of ‘I don’t vouch for this information.’ Such exten- verbs. The verb of ‘saying’ is a frequent source for sions are not universal. As Valenzuela (2003: 57) reported and quotative evidentials, and the verb ‘feel, remarks for Shipibo-Konibo, the selection of reported think, hear’ can give rise to a nonvisual evidential in evidential over the direct evidential ‘‘does not indicate large systems. Closed word classes – deictics and uncertainty or a lesser degree of reliability but simply locatives – may give rise to evidentials, both in small reported information.’’ and in large systems. Languages with multiterm evidentials generally Evidentiality strategies involving past tenses and tend to have a multiplicity of other verbal categories, perfects, and nominalizations, can develop into especially ones that relate to modalities. The larger small evidentiality systems (A1 and A2). The creation the evidential system, the less likely it is that the of a reported evidential may involve reanalysis of evidential terms will develop epistemic extensions. subordinate clauses (typically, complement clauses A nonfirsthand term in a two-term system, or an of verbs of speech) as main clauses (as in Estonian). inferred term in a three-term system, tend to subsume Nonindicative moods and modalities may give rise to all sorts of information acquired indirectly. These a term in a large evidentiality system; however, there evidentials may then evolve mirative extensions are no examples of a modal system developing into a (to do with unexpected information, the ‘unprepared system of evidentials. This lack of evidence confirms mind’ of the speaker, and speaker’s surprise: the separate status of evidentiality and modality. DeLancey, 1997, 2001). Large evidential systems tend to be heterogenous in When used with a first person , the nonfirst- origin. hand evidentials in A1 and A2 systems, nonvisual Evidentiality is a property of a significant number evidentials in larger systems, and reported in sys- of linguistic areas, including the Balkans, the Baltic tems of varied types may acquire additional mean- area, India, and a variety of locations in Amazonia ings of lack of intention, control, awareness, and (Aikhenvald and Dixon, 1998). Evidentials may on the part of the speaker. Verbs covering make their way into contact languages, such as internal states may require obligatory evidential (see papers in Hardman, 1981). choice depending on person. As a result of these If several information sources are available – for correlations evidentials acquire the implicit value of instance, I both saw and heard a dog barking and person markers. later someone told me about it – any one of three Evidentials interrelate with clause types and other evidentials can potentially be used: visual, nonvisual, grammatical categories in the following ways: and reported. In this situation, the visual evidential 1. The maximum number of evidential specifications tends to be preferred. The genre of a text may deter- tends to be distinguished in declarative main mine the choice of an evidential. Traditional stories clauses. are typically cast in reported evidential. Evidentials Evidentiality in Grammar 325 can be manipulated in as a stylistic device. Dendale P (1993). ‘Le conditionnel de l’information incer- Switching from a reported to a direct (or visual) evi- taine: marqueur modal ou marqueur evidentiel?’ In Hilty dential creates the effect of the speaker’s participation G (ed.) Proceedings of the XXe Congre`s International de and confidence. Switching to a nonfirsthand eviden- Linguistique et Philologie Romanes, Tome I, Section I. La tial often implies a backgrounded aside. Evidentiality phrase. Tu¨ bingen: Francke. 165–176. Dickinson C (2000). ‘ in Tsafiki.’ Studies in Lan- is interlinked with conventionalized attitudes to in- guage 24, 379–421. formation and precision in stating the source of Dixon R M W (2003). ‘Evidentiality in Jarawara.’ In information (Hardman, 1981, 1986). Aikhenvald & Dixon (eds.). 165–188. Floyd R (1999). The structure of evidential categories in See also: ArawakLanguages; Epistemology and Lan- Wanka Quechua. Arlington: Summer Institute of Lin- guage; Inference: Abduction, Induction, Deduction; guistics /University of Texas. 165–188. Mood and Modality in Grammar; Reported Speech: Prag- Guentche´va Z (ed.) (1996). L’E´ nonciation me´diatise´e. Lou- matic Aspects; Tense, Mood, Aspect: Overview. vain-Paris: E´ ditions Peeters. Hansson I-L (2003). ‘Akha.’ In Thurgood G & LaPolla R J (eds.) The Sino-Tibetan languages. London: Routledge. Bibliography 236–252. Hardman M J (ed.) (1981). The in its Aikhenvald A Y (2003a). ‘Evidentiality in typological per- social and cultural context: a collection of essays on spective.’ In Aikhenvald & Dixon (eds.). 1–31. aspects of Aymara language and culture. Gainesville: Aikhenvald A Y (2003b). ‘Evidentiality in Tariana.’ In University Presses of Florida. Aikhenvald & Dixon (eds.). 131–164. Hardman M J (1986). ‘Data-source marking in the Jaqi Aikhenvald AY (2004). Evidentiality. Oxford: Oxford Uni- languages.’ In Chafe & Nichols (eds.). 113–136. versity Press. Jacobsen Jr & William H (1986). ‘The heterogeneity Aikhenvald A Y & Dixon R M W (1998). ‘Evidentials and of evidentials in Makah.’ In Chafe & Nichols (eds.). areal typology: a case-study from Amazonia.’ Language 3–28. Sciences 20, 241–257. Jakobson R O (1957). Shifters, verbal categories, and the Aikhenvald A Y & Dixon R M W (eds.) (2003). Studies in Russian verb. Cambridge: Harvard University. evidentiality. Amsterdam: John Benjamins. Johanson L (2003). ‘Evidentiality in Turkic.’ In Aikhenvald Barnes J (1984). ‘Evidentials in the Tuyuca verb.’ Interna- & Dixon (eds.). 273–291. tional Journal of American Linguistics 50, 255–271. Johanson L & Utas B (eds.) (2000). Evidentials. Turkic, Boas F (1938). ‘Language.’ In Boas F (ed.) General Anthro- Iranian and neighbouring languages. Berlin: Mouton de pology. Boston/New York: D. C. Heath and Company. Gruyter. 124–145. Lazard G (1999). ‘Mirativity, evidentiality, mediativity, or Chafe W L & Nichols J (eds.) (1986). Evidentiality: the other?’ 3, 91–110. linguistic coding of epistemology. Norwood, NJ: Ablex. Lazard G (2001). ‘On the of evidenti- Chirikba V (2003). ‘Evidential category and evidential ality.’ Journal of 33, 358–368. strategy in Abkhaz.’ In Aikhenvald & Dixon (eds.). Palmer F R (1986). Mood and modality. Cambridge: Cam- 243–272. bridge University Press. De Haan F (1999). ‘Evidentiality and : de Reuse W J (2003). ‘Evidentiality in Western Apache.’ In setting boundaries.’ Southwest Journal of Linguistics 18, Aikhenvald & Dixon (eds.). 79–100. 83–102. Silver S & Miller W (1997). American Indian languages. DeLancey S (1997). ‘Mirativity: the grammatical marking cultural and social contexts. Tucson: The University of of unexpected information.’ Linguistic Typology 1, Arizona Press. 33–52. Valenzuela P (2003). ‘Evidentiality in Shipibo-Konibo, with DeLancey S (2001). ‘The mirative and evidentiality.’ Jour- a comparative overview of the category in Panoan.’ In nal of Pragmatics 33, 369–382. Aikhenvald & Dixon (eds.) 33–62.